IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

DOCS shift to rst - Automatic Speech Recognition C++ Sample & Automatic Speech Recognition Python* Sample (#16609)

Browse Source
This commit is contained in:
Maciej Smyk
2023-03-29 16:39:09 +02:00
committed by GitHub
parent 10668f4f3a
commit 7d8f4af78a
2 changed files with 508 additions and 451 deletions
Download Patch File Download Diff File Expand all files Collapse all files
samples/cpp/speech_sample/README.md
+126 -85
View File
@@ -1,94 +1,121 @@
# Automatic Speech Recognition C++ Sample {#openvino_inference_engine_samples_speech_sample_README}
@sphinxdirective
This sample demonstrates how to execute an Asynchronous Inference of acoustic model based on Kaldi\* neural networks and speech feature vectors.
The sample works with Kaldi ARK or Numpy* uncompressed NPZ files, so it does not cover an end-to-end speech recognition scenario (speech to text), requiring additional preprocessing (feature extraction) to get a feature vector from a speech signal, as well as postprocessing (decoding) to produce text from scores.
The following C++ API is used in the application:
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Feature | API | Description |
| :--- | :--- | :--- |
| Available Devices | `ov::Core::get_available_devices`, `ov::Core::get_property` | Get information of the devices for inference |
| Import/Export Model | `ov::Core::import_model`, `ov::CompiledModel::export_model` | The GNA plugin supports loading and saving of the GNA-optimized model |
| Model Operations | `ov::set_batch`, `ov::Model::add_output`, `ov::CompiledModel::inputs`, `ov::CompiledModel::outputs` | Managing of model: configure batch_size, input and output tensors |
| Node Operations | `ov::OutputVector::size`, `ov::Output::get_shape` | Get node shape |
| Asynchronous Infer | `ov::InferRequest::start_async`, `ov::InferRequest::wait` | Do asynchronous inference and waits until inference result becomes available |
| InferRequest Operations | `ov::InferRequest::query_state`, `ov::VariableState::reset` | Gets and resets CompiledModel state control |
| Tensor Operations | `ov::Tensor::get_size`, `ov::Tensor::data`, `ov::InferRequest::get_tensor` | Get a tensor, its size and data |
| Profiling | `ov::InferRequest::get_profiling_info` | Get infer request profiling info |
+=============================================================+=============================================================================================================+==============================================================================+
| Available Devices | ``ov::Core::get_available_devices``, ``ov::Core::get_property`` | Get information of the devices for inference |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Import/Export Model | ``ov::Core::import_model``, ``ov::CompiledModel::export_model`` | The GNA plugin supports loading and saving of the GNA-optimized model |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Model Operations | ``ov::set_batch``, ``ov::Model::add_output``, ``ov::CompiledModel::inputs``, ``ov::CompiledModel::outputs`` | Managing of model: configure batch_size, input and output tensors |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Node Operations | ``ov::OutputVector::size``, ``ov::Output::get_shape`` | Get node shape |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Asynchronous Infer | ``ov::InferRequest::start_async``, ``ov::InferRequest::wait`` | Do asynchronous inference and waits until inference result becomes available |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| InferRequest Operations | ``ov::InferRequest::query_state``, ``ov::VariableState::reset`` | Gets and resets CompiledModel state control |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Tensor Operations | ``ov::Tensor::get_size``, ``ov::Tensor::data``, ``ov::InferRequest::get_tensor`` | Get a tensor, its size and data |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
| Profiling | ``ov::InferRequest::get_profiling_info`` | Get infer request profiling info |
+-------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------+------------------------------------------------------------------------------+
Basic OpenVINO™ Runtime API is covered by [Hello Classification C++ sample](../hello_classification/README.md).
Basic OpenVINO™ Runtime API is covered by :doc:`Hello Classification C++ sample <openvino_inference_engine_samples_hello_classification_README>`.
+-------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Options | Values |
| :--- | :--- |
| Validated Models | Acoustic model based on Kaldi\* neural networks (see [Model Preparation](#model-preparation-speech) section) |
+=============================================================+===============================================================================================================================================================+
| Validated Models | Acoustic model based on Kaldi\* neural networks (see :ref:`Model Preparation <model-preparation-speech>` section) |
+-------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Model Format | OpenVINO™ toolkit Intermediate Representation (\*.xml + \*.bin) |
| Supported devices | See [Execution Modes](#execution-modes-speech) section below and [List Supported Devices](../../../docs/OV_Runtime_UG/supported_plugins/Supported_Devices.md) |
+-------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Supported devices | See :ref:`Execution Modes <execution-modes-speech>` section below and :doc:`List Supported Devices <openvino_docs_OV_UG_supported_plugins_Supported_Devices>` |
+-------------------------------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
## How It Works
How It Works
############
At startup, the sample application reads command-line parameters, loads a specified model and input data to the OpenVINO™ Runtime plugin, performs inference on all speech utterances stored in the input file(s), logging each step in a standard output stream.
If the `-r` option is given, error statistics are provided for each speech utterance as shown above.
If the ``-r`` option is given, error statistics are provided for each speech utterance as shown above.
You can see the explicit description of
each sample step at [Integration Steps](../../../docs/OV_Runtime_UG/integrate_with_your_application.md) section of "Integrate OpenVINO™ Runtime with Your Application" guide.
each sample step at :doc:`Integration Steps <openvino_docs_OV_UG_Integrate_OV_with_your_application>` section of "Integrate OpenVINO™ Runtime with Your Application" guide.
### GNA-specific details
GNA-specific details
++++++++++++++++++++
#### Quantization
Quantization
------------
If the GNA device is selected (for example, using the `-d` GNA flag), the GNA OpenVINO™ Runtime plugin quantizes the model and input feature vector sequence to integer representation before performing inference.
Several parameters control neural network quantization. The `-q` flag determines the quantization mode.
If the GNA device is selected (for example, using the ``-d`` GNA flag), the GNA OpenVINO™ Runtime plugin quantizes the model and input feature vector sequence to integer representation before performing inference.
Several parameters control neural network quantization. The ``-q`` flag determines the quantization mode.
Two modes are supported:
- *static* - The first utterance in the input file is scanned for dynamic range. The scale factor (floating point scalar multiplier) required to scale the maximum input value of the first utterance to 16384 (15 bits) is used for all subsequent inputs. The neural network is quantized to accommodate the scaled input dynamic range.
- *user-defined* - The user may specify a scale factor via the `-sf` flag that will be used for static quantization.
- *user-defined* - The user may specify a scale factor via the ``-sf`` flag that will be used for static quantization.
The `-qb` flag provides a hint to the GNA plugin regarding the preferred target weight resolution for all layers. For example, when `-qb 8` is specified, the plugin will use 8-bit weights wherever possible in the
The ``-qb`` flag provides a hint to the GNA plugin regarding the preferred target weight resolution for all layers. For example, when ``-qb 8`` is specified, the plugin will use 8-bit weights wherever possible in the
network.
> **NOTE**:
>
> - It is not always possible to use 8-bit weights due to GNA hardware limitations. For example, convolutional layers always use 16-bit weights (GNA hardware version 1 and 2). This limitation will be removed in GNA hardware version 3 and higher.
.. note::
#### <a name="execution-modes-speech"></a> Execution Modes
It is not always possible to use 8-bit weights due to GNA hardware limitations. For example, convolutional layers always use 16-bit weights (GNA hardware version 1 and 2). This limitation will be removed in GNA hardware version 3 and higher.
Several execution modes are supported via the `-d` flag:
- `CPU` - All calculations are performed on CPU device using CPU Plugin.
- `GPU` - All calculations are performed on GPU device using GPU Plugin.
- `VPUX` - All calculations are performed on VPUX device using VPUX Plugin.
- `GNA_AUTO` - GNA hardware is used if available and the driver is installed. Otherwise, the GNA device is emulated in fast-but-not-bit-exact mode.
- `GNA_HW` - GNA hardware is used if available and the driver is installed. Otherwise, an error will occur.
- `GNA_SW` - Deprecated. The GNA device is emulated in fast-but-not-bit-exact mode.
- `GNA_SW_FP32` - Substitutes parameters and calculations from low precision to floating point (FP32).
- `GNA_SW_EXACT` - GNA device is emulated in bit-exact mode.
.. _execution-modes-speech:
#### Loading and Saving Models
Execution Modes
---------------
The GNA plugin supports loading and saving of the GNA-optimized model (non-IR) via the `-rg` and `-wg` flags. Thereby, it is possible to avoid the cost of full model quantization at run time. The GNA plugin also supports export of firmware-compatible embedded model images for the Intel® Speech Enabling Developer Kit and Amazon Alexa* Premium Far-Field Voice Development Kit via the `-we` flag (save only).
Several execution modes are supported via the ``-d`` flag:
- ``CPU`` - All calculations are performed on CPU device using CPU Plugin.
- ``GPU`` - All calculations are performed on GPU device using GPU Plugin.
- ``VPUX`` - All calculations are performed on VPUX device using VPUX Plugin.
- ``GNA_AUTO`` - GNA hardware is used if available and the driver is installed. Otherwise, the GNA device is emulated in fast-but-not-bit-exact mode.
- ``GNA_HW`` - GNA hardware is used if available and the driver is installed. Otherwise, an error will occur.
- ``GNA_SW`` - Deprecated. The GNA device is emulated in fast-but-not-bit-exact mode.
- ``GNA_SW_FP32`` - Substitutes parameters and calculations from low precision to floating point (FP32).
- ``GNA_SW_EXACT`` - GNA device is emulated in bit-exact mode.
Loading and Saving Models
-------------------------
The GNA plugin supports loading and saving of the GNA-optimized model (non-IR) via the ``-rg`` and ``-wg`` flags. Thereby, it is possible to avoid the cost of full model quantization at run time. The GNA plugin also supports export of firmware-compatible embedded model images for the Intel® Speech Enabling Developer Kit and Amazon Alexa* Premium Far-Field Voice Development Kit via the ``-we`` flag (save only).
In addition to performing inference directly from a GNA model file, these combinations of options make it possible to:
- Convert from IR format to GNA format model file (`-m`, `-wg`)
- Convert from IR format to embedded format model file (`-m`, `-we`)
- Convert from GNA format to embedded format model file (`-rg`, `-we`)
- Convert from IR format to GNA format model file (``-m``, ``-wg``)
- Convert from IR format to embedded format model file (``-m``, ``-we``)
- Convert from GNA format to embedded format model file (``-rg``, ``-we``)
## Building
Building
########
To build the sample, please use instructions available at [Build the Sample Applications](../../../docs/OV_Runtime_UG/Samples_Overview.md) section in OpenVINO™ Toolkit Samples guide.
To build the sample, please use instructions available at :doc:`Build the Sample Applications <openvino_docs_OV_UG_Samples_Overview>` section in OpenVINO™ Toolkit Samples guide.
## Running
Running
#######
Run the application with the -h option to see the usage message:
```
.. code-block:: sh
speech_sample -h
```
Usage message:
```
.. code-block:: sh
[ INFO ] OpenVINO Runtime version ......... <version>
[ INFO ] Build ........... <build>
[ INFO ]
@@ -120,15 +147,18 @@ Options:
-memory_reuse_off Optional. Disables memory optimizations for compiled model.
Available target devices: CPU GNA GPU VPUX
```
### <a name="model-preparation-speech"></a> Model Preparation
.. _model-preparation-speech:
Model Preparation
+++++++++++++++++
You can use the following model optimizer command to convert a Kaldi nnet1 or nnet2 neural model to OpenVINO™ toolkit Intermediate Representation format:
```
.. code-block:: sh
mo --framework kaldi --input_model wsj_dnn5b.nnet --counts wsj_dnn5b.counts --remove_output_softmax --output_dir <OUTPUT_MODEL_DIR>
```
The following pre-trained models are available:
@@ -136,29 +166,32 @@ The following pre-trained models are available:
- rm_lstm4f
- wsj_dnn5b_smbr
All of them can be downloaded from [https://storage.openvinotoolkit.org/models_contrib/speech/2021.2](https://storage.openvinotoolkit.org/models_contrib/speech/2021.2).
All of them can be downloaded from `the storage <https://storage.openvinotoolkit.org/models_contrib/speech/2021.2>`__.
### Speech Inference
Speech Inference
++++++++++++++++
Once the IR is created, you can do inference on Intel® Processors with the GNA co-processor (or emulation library):
```
.. code-block:: sh
speech_sample -m wsj_dnn5b.xml -i dev93_10.ark -r dev93_scores_10.ark -d GNA_AUTO -o result.ark
```
Here, the floating point Kaldi-generated reference neural network scores (`dev93_scores_10.ark`) corresponding to the input feature file (`dev93_10.ark`) are assumed to be available for comparison.
Here, the floating point Kaldi-generated reference neural network scores (``dev93_scores_10.ark``) corresponding to the input feature file (``dev93_10.ark``) are assumed to be available for comparison.
> **NOTES**:
>
> - Before running the sample with a trained model, make sure the model is converted to the intermediate representation (IR) format (\*.xml + \*.bin) using the [Model Optimizer tool](../../../docs/MO_DG/Deep_Learning_Model_Optimizer_DevGuide.md).
>
> - The sample supports input and output in numpy file format (.npz)
.. note::
## Sample Output
- Before running the sample with a trained model, make sure the model is converted to the intermediate representation (IR) format (\*.xml + \*.bin) using the :doc:`Model Optimizer tool <openvino_docs_MO_DG_Deep_Learning_Model_Optimizer_DevGuide>`.
- The sample supports input and output in numpy file format (.npz)
Sample Output
#############
The sample application logs each step in a standard output stream.
```
.. code-block:: sh
[ INFO ] OpenVINO runtime: OpenVINO Runtime version ......... 2022.1.0
[ INFO ] Build ........... 2022.1.0-6311-a90bb1ff017
[ INFO ]
@@ -199,40 +232,48 @@ End of Utterance 1
End of Utterance X
[ INFO ] Execution successful
```
## Use of Sample in Kaldi* Speech Recognition Pipeline
Use of Sample in Kaldi* Speech Recognition Pipeline
###################################################
The Wall Street Journal DNN model used in this example was prepared using the Kaldi s5 recipe and the Kaldi Nnet (nnet1) framework. It is possible to recognize speech by substituting the `speech_sample` for
Kaldi's nnet-forward command. Since the `speech_sample` does not yet use pipes, it is necessary to use temporary files for speaker-transformed feature vectors and scores when running the Kaldi speech recognition pipeline. The following operations assume that feature extraction was already performed according to the `s5` recipe and that the working directory within the Kaldi source tree is `egs/wsj/s5`.
The Wall Street Journal DNN model used in this example was prepared using the Kaldi s5 recipe and the Kaldi Nnet (nnet1) framework. It is possible to recognize speech by substituting the ``speech_sample`` for
Kaldi's nnet-forward command. Since the ``speech_sample`` does not yet use pipes, it is necessary to use temporary files for speaker-transformed feature vectors and scores when running the Kaldi speech recognition pipeline. The following operations assume that feature extraction was already performed according to the ``s5`` recipe and that the working directory within the Kaldi source tree is ``egs/wsj/s5``.
1. Prepare a speaker-transformed feature set given the feature transform specified in ``final.feature_transform`` and the feature files specified in ``feats.scp``:
.. code-block:: sh
1. Prepare a speaker-transformed feature set given the feature transform specified in `final.feature_transform` and the feature files specified in `feats.scp`:
```sh
nnet-forward --use-gpu=no final.feature_transform "ark,s,cs:copy-feats scp:feats.scp ark:- |" ark:feat.ark
```
2. Score the feature set using the `speech_sample`:
```sh
2. Score the feature set using the ``speech_sample``:
.. code-block:: sh
./speech_sample -d GNA_AUTO -bs 8 -i feat.ark -m wsj_dnn5b.xml -o scores.ark
```
OpenVINO™ toolkit Intermediate Representation `wsj_dnn5b.xml` file was generated in the previous [Model Preparation](#model-preparation-speech) section.
OpenVINO™ toolkit Intermediate Representation ``wsj_dnn5b.xml`` file was generated in the previous :ref:`Model Preparation <model-preparation-speech>` section.
3. Run the Kaldi decoder to produce n-best text hypotheses and select most likely text given the WFST (``HCLG.fst``), vocabulary (``words.txt``), and TID/PID mapping (``final.mdl``):
.. code-block:: sh
3. Run the Kaldi decoder to produce n-best text hypotheses and select most likely text given the WFST (`HCLG.fst`), vocabulary (`words.txt`), and TID/PID mapping (`final.mdl`):
```sh
latgen-faster-mapped --max-active=7000 --max-mem=50000000 --beam=13.0 --lattice-beam=6.0 --acoustic-scale=0.0833 --allow-partial=true --word-symbol-table=words.txt final.mdl HCLG.fst ark:scores.ark ark:-| lattice-scale --inv-acoustic-scale=13 ark:- ark:- | lattice-best-path --word-symbol-table=words.txt ark:- ark,t:- > out.txt &
```
4. Run the word error rate tool to check accuracy given the vocabulary (`words.txt`) and reference transcript (`test_filt.txt`):
```sh
4. Run the word error rate tool to check accuracy given the vocabulary (``words.txt``) and reference transcript (``test_filt.txt``):
.. code-block:: sh
cat out.txt | utils/int2sym.pl -f 2- words.txt | sed s:\<UNK\>::g | compute-wer --text --mode=present ark:test_filt.txt ark,p:-
```
All of mentioned files can be downloaded from [https://storage.openvinotoolkit.org/models_contrib/speech/2021.2/wsj_dnn5b_smbr](https://storage.openvinotoolkit.org/models_contrib/speech/2021.2/wsj_dnn5b_smbr)
All of mentioned files can be downloaded from `the storage <https://storage.openvinotoolkit.org/models_contrib/speech/2021.2/wsj_dnn5b_smbr>`__
## See Also
See Also
########
- :doc:`Integrate the OpenVINO™ Runtime with Your Application <openvino_docs_OV_UG_Integrate_OV_with_your_application>`
- :doc:`Using OpenVINO™ Toolkit Samples <openvino_docs_OV_UG_Samples_Overview>`
- :doc:`Model Downloader <omz_tools_downloader>`
- :doc:`Model Optimizer <openvino_docs_MO_DG_Deep_Learning_Model_Optimizer_DevGuide>`
@endsphinxdirective
- [Integrate the OpenVINO™ Runtime with Your Application](../../../docs/OV_Runtime_UG/integrate_with_your_application.md)
- [Using OpenVINO™ Toolkit Samples](../../../docs/OV_Runtime_UG/Samples_Overview.md)
- [Model Downloader](@ref omz_tools_downloader)
- [Model Optimizer](../../../docs/MO_DG/Deep_Learning_Model_Optimizer_DevGuide.md)
samples/python/speech_sample/README.md
+98 -82
View File
@@ -1,92 +1,115 @@
# Automatic Speech Recognition Python* Sample {#openvino_inference_engine_ie_bridges_python_sample_speech_sample_README}
@sphinxdirective
This sample demonstrates how to do a Synchronous Inference of acoustic model based on Kaldi\* neural models and speech feature vectors.
The sample works with Kaldi ARK or Numpy* uncompressed NPZ files, so it does not cover an end-to-end speech recognition scenario (speech to text), requiring additional preprocessing (feature extraction) to get a feature vector from a speech signal, as well as postprocessing (decoding) to produce text from scores.
Automatic Speech Recognition Python sample application demonstrates how to use the following Python API in applications:
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
| Feature | API | Description |
| :---------------------- | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :-------------------------------------------------------------------- |
| Import/Export Model | [openvino.runtime.Core.import_model], [openvino.runtime.CompiledModel.export_model] | The GNA plugin supports loading and saving of the GNA-optimized model |
| Model Operations | [openvino.runtime.Model.add_outputs], [openvino.runtime.set_batch], [openvino.runtime.CompiledModel.inputs], [openvino.runtime.CompiledModel.outputs], [openvino.runtime.ConstOutput.any_name] | Managing of model: configure batch_size, input and output tensors |
| Synchronous Infer | [openvino.runtime.CompiledModel.create_infer_request], [openvino.runtime.InferRequest.infer] | Do synchronous inference |
| InferRequest Operations | [openvino.runtime.InferRequest.get_input_tensor], [openvino.runtime.InferRequest.model_outputs], [openvino.runtime.InferRequest.model_inputs], | Get info about model using infer request API |
| InferRequest Operations | [openvino.runtime.InferRequest.query_state], [openvino.runtime.VariableState.reset] | Gets and resets CompiledModel state control |
| Profiling | [openvino.runtime.InferRequest.profiling_info], [openvino.runtime.ProfilingInfo.real_time] | Get infer request profiling info |
+===================================================================+================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================================+=======================================================================+
| Import/Export Model | `openvino.runtime.Core.import_model <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.Core.html#openvino.runtime.Core.import_model>`__ , `openvino.runtime.CompiledModel.export_model <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.export_model>`__ | The GNA plugin supports loading and saving of the GNA-optimized model |
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
| Model Operations | `openvino.runtime.Model.add_outputs <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.Model.html#openvino.runtime.Model.add_outputs>`__ , `openvino.runtime.set_batch <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.html#openvino.runtime.set_batch>`__ , `openvino.runtime.CompiledModel.inputs <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.inputs>`__ , `openvino.runtime.CompiledModel.outputs <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.outputs>`__ , `openvino.runtime.ConstOutput.any_name <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.ConstOutput.html#openvino.runtime.ConstOutput.any_name>`__ | Managing of model: configure batch_size, input and output tensors |
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
| Synchronous Infer | `openvino.runtime.CompiledModel.create_infer_request <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.create_infer_request>`__ , `openvino.runtime.InferRequest.infer <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.infer>`__ | Do synchronous inference |
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
| InferRequest Operations | `openvino.runtime.InferRequest.get_input_tensor <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.get_input_tensor>`__ , `openvino.runtime.InferRequest.model_outputs <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.model_outputs>`__ , `openvino.runtime.InferRequest.model_inputs <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.model_inputs>`__ , | Get info about model using infer request API |
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
| InferRequest Operations | `openvino.runtime.InferRequest.query_state <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.query_state>`__ , `openvino.runtime.VariableState.reset <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.inference_engine.VariableState.html#openvino.inference_engine.VariableState.reset>`__ | Gets and resets CompiledModel state control |
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
| Profiling | `openvino.runtime.InferRequest.profiling_info <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.profiling_info>`__ , `openvino.runtime.ProfilingInfo.real_time <https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.ProfilingInfo.html#openvino.runtime.ProfilingInfo.real_time>`__ | Get infer request profiling info |
+-------------------------------------------------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-----------------------------------------------------------------------+
Basic OpenVINO™ Runtime API is covered by [Hello Classification Python* Sample](../hello_classification/README.md).
Basic OpenVINO™ Runtime API is covered by :doc:`Hello Classification Python* Sample <openvino_inference_engine_ie_bridges_python_sample_hello_classification_README>`.
+----------------------------------------------------------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Options | Values |
| :------------------------- | :--------------------------------------------------------------------------------------------------------------------------------------------- |
| Validated Models | Acoustic model based on Kaldi* neural models (see [Model Preparation](#model-preparation-speech-python) section) |
+======================================================================+======================================================================================================================================================================+
| Validated Models | Acoustic model based on Kaldi* neural models (see :ref:`Model Preparation <model-preparation-speech-python>` section) |
+----------------------------------------------------------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Model Format | OpenVINO™ toolkit Intermediate Representation (.xml + .bin) |
| Supported devices | See [Execution Modes](#execution-modes-speech-python) section below and [List Supported Devices](../../../docs/OV_Runtime_UG/supported_plugins/Supported_Devices.md) |
| Other language realization | [C++](../../../samples/cpp/speech_sample/README.md) |
+----------------------------------------------------------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Supported devices | See :ref:`Execution Modes <execution-modes-speech-python>` section below and :doc:`List Supported Devices <openvino_docs_OV_UG_supported_plugins_Supported_Devices>` |
+----------------------------------------------------------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Other language realization | :doc:`C++ <openvino_inference_engine_samples_speech_sample_README>` |
+----------------------------------------------------------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
## How It Works
How It Works
############
At startup, the sample application reads command-line parameters, loads a specified model and input data to the OpenVINO™ Runtime plugin, performs synchronous inference on all speech utterances stored in the input file, logging each step in a standard output stream.
You can see the explicit description of
each sample step at [Integration Steps](../../../docs/OV_Runtime_UG/integrate_with_your_application.md) section of "Integrate OpenVINO™ Runtime with Your Application" guide.
each sample step at :doc:`Integration Steps <openvino_docs_OV_UG_Integrate_OV_with_your_application>` section of "Integrate OpenVINO™ Runtime with Your Application" guide.
## GNA-specific details
GNA-specific details
####################
### Quantization
Quantization
++++++++++++
If the GNA device is selected (for example, using the `-d` GNA flag), the GNA OpenVINO™ Runtime plugin quantizes the model and input feature vector sequence to integer representation before performing inference.
If the GNA device is selected (for example, using the ``-d`` GNA flag), the GNA OpenVINO™ Runtime plugin quantizes the model and input feature vector sequence to integer representation before performing inference.
Several neural model quantization modes:
- *static* - The first utterance in the input file is scanned for dynamic range. The scale factor (floating point scalar multiplier) required to scale the maximum input value of the first utterance to 16384 (15 bits) is used for all subsequent inputs. The neural model is quantized to accommodate the scaled input dynamic range.
- *user-defined* - The user may specify a scale factor via the `-sf` flag that will be used for static quantization.
- *user-defined* - The user may specify a scale factor via the ``-sf`` flag that will be used for static quantization.
The `-qb` flag provides a hint to the GNA plugin regarding the preferred target weight resolution for all layers.
For example, when `-qb 8` is specified, the plugin will use 8-bit weights wherever possible in the
The ``-qb`` flag provides a hint to the GNA plugin regarding the preferred target weight resolution for all layers.
For example, when ``-qb 8`` is specified, the plugin will use 8-bit weights wherever possible in the
model.
> **NOTE**:
>
> - It is not always possible to use 8-bit weights due to GNA hardware limitations. For example, convolutional layers always use 16-bit weights (GNA hardware version 1 and 2). This limitation will be removed in GNA hardware version 3 and higher.
>
.. note::
### <a name="execution-modes-speech-python"></a> Execution Modes
It is not always possible to use 8-bit weights due to GNA hardware limitations. For example, convolutional layers always use 16-bit weights (GNA hardware version 1 and 2). This limitation will be removed in GNA hardware version 3 and higher.
Several execution modes are supported via the `-d` flag:
.. _execution-modes-speech-python:
- `CPU` - All calculations are performed on CPU device using CPU Plugin.
- `GPU` - All calculations are performed on GPU device using GPU Plugin.
- `VPUX` - All calculations are performed on VPUX device using VPUX Plugin.
- `GNA_AUTO` - GNA hardware is used if available and the driver is installed. Otherwise, the GNA device is emulated in fast-but-not-bit-exact mode.
- `GNA_HW` - GNA hardware is used if available and the driver is installed. Otherwise, an error will occur.
- `GNA_SW` - Deprecated. The GNA device is emulated in fast-but-not-bit-exact mode.
- `GNA_SW_FP32` - Substitutes parameters and calculations from low precision to floating point (FP32).
- `GNA_SW_EXACT` - GNA device is emulated in bit-exact mode.
Execution Modes
+++++++++++++++
### Loading and Saving Models
Several execution modes are supported via the ``-d`` flag:
The GNA plugin supports loading and saving of the GNA-optimized model (non-IR) via the `-rg` and `-wg` flags.
- ``CPU`` - All calculations are performed on CPU device using CPU Plugin.
- ``GPU`` - All calculations are performed on GPU device using GPU Plugin.
- ``VPUX`` - All calculations are performed on VPUX device using VPUX Plugin.
- ``GNA_AUTO`` - GNA hardware is used if available and the driver is installed. Otherwise, the GNA device is emulated in fast-but-not-bit-exact mode.
- ``GNA_HW`` - GNA hardware is used if available and the driver is installed. Otherwise, an error will occur.
- ``GNA_SW`` - Deprecated. The GNA device is emulated in fast-but-not-bit-exact mode.
- ``GNA_SW_FP32`` - Substitutes parameters and calculations from low precision to floating point (FP32).
- ``GNA_SW_EXACT`` - GNA device is emulated in bit-exact mode.
Loading and Saving Models
+++++++++++++++++++++++++
The GNA plugin supports loading and saving of the GNA-optimized model (non-IR) via the ``-rg`` and ``-wg`` flags.
Thereby, it is possible to avoid the cost of full model quantization at run time.
The GNA plugin also supports export of firmware-compatible embedded model images for the Intel® Speech Enabling Developer Kit and Amazon Alexa* Premium Far-Field Voice Development Kit via the `-we` flag (save only).
The GNA plugin also supports export of firmware-compatible embedded model images for the Intel® Speech Enabling Developer Kit and Amazon Alexa* Premium Far-Field Voice Development Kit via the ``-we`` flag (save only).
In addition to performing inference directly from a GNA model file, these options make it possible to:
- Convert from IR format to GNA format model file (`-m`, `-wg`)
- Convert from IR format to embedded format model file (`-m`, `-we`)
- Convert from GNA format to embedded format model file (`-rg`, `-we`)
- Convert from IR format to GNA format model file (``-m``, ``-wg``)
- Convert from IR format to embedded format model file (``-m``, ``-we``)
- Convert from GNA format to embedded format model file (``-rg``, ``-we``)
## Running
Running
#######
Run the application with the `-h` option to see the usage message:
Run the application with the ``-h`` option to see the usage message:
.. code-block:: bash
```
python speech_sample.py -h
```
Usage message:
```
.. code-block:: console
usage: speech_sample.py [-h] (-m MODEL | -rg IMPORT_GNA_MODEL) -i INPUT [-o OUTPUT] [-r REFERENCE] [-d DEVICE] [-bs [1-8]]
[-layout LAYOUT] [-qb [8, 16]] [-sf SCALE_FACTOR] [-wg EXPORT_GNA_MODEL]
[-we EXPORT_EMBEDDED_GNA_MODEL] [-we_gen [GNA1, GNA3]]
@@ -150,15 +173,18 @@ Options:
models. If you use the cw_l or cw_r flag, then batch size argument is ignored.
-pwl_me PWL_ME Optional. The maximum percent of error for PWL function. The value must be in <0, 100> range. The
default value is 1.0.
```
## <a name="model-preparation-speech-python"></a> Model Preparation
.. _model-preparation-speech-python:
Model Preparation
#################
You can use the following model optimizer command to convert a Kaldi nnet1 or nnet2 neural model to OpenVINO™ toolkit Intermediate Representation format:
```sh
.. code-block:: sh
mo --framework kaldi --input_model wsj_dnn5b.nnet --counts wsj_dnn5b.counts --remove_output_softmax --output_dir <OUTPUT_MODEL_DIR>
```
The following pre-trained models are available:
@@ -166,27 +192,30 @@ The following pre-trained models are available:
- rm_lstm4f
- wsj_dnn5b_smbr
All of them can be downloaded from [https://storage.openvinotoolkit.org/models_contrib/speech/2021.2](https://storage.openvinotoolkit.org/models_contrib/speech/2021.2).
All of them can be downloaded from `the storage <https://storage.openvinotoolkit.org/models_contrib/speech/2021.2>`.
## Speech Inference
Speech Inference
################
You can do inference on Intel® Processors with the GNA co-processor (or emulation library):
```
.. code-block:: sh
python speech_sample.py -m wsj_dnn5b.xml -i dev93_10.ark -r dev93_scores_10.ark -d GNA_AUTO -o result.npz
```
> **NOTES**:
>
> - Before running the sample with a trained model, make sure the model is converted to the intermediate representation (IR) format (\*.xml + \*.bin) using the [Model Optimizer tool](../../../docs/MO_DG/Deep_Learning_Model_Optimizer_DevGuide.md).
>
> - The sample supports input and output in numpy file format (.npz)
## Sample Output
.. note::
- Before running the sample with a trained model, make sure the model is converted to the intermediate representation (IR) format (\*.xml + \*.bin) using the :doc:`Model Optimizer tool <openvino_docs_MO_DG_Deep_Learning_Model_Optimizer_DevGuide>`.
- The sample supports input and output in numpy file format (.npz)
Sample Output
#############
The sample application logs each step in a standard output stream.
```
.. code-block:: sh
[ INFO ] Creating OpenVINO Runtime Core
[ INFO ] Reading the model: /models/wsj_dnn5b_smbr_fp32.xml
[ INFO ] Using scale factor(s) calculated from first utterance
@@ -327,28 +356,15 @@ The sample application logs each step in a standard output stream.
[ INFO ] Total sample time: 39722.60ms
[ INFO ] File result.npz was created!
[ INFO ] This sample is an API example, for any performance measurements please use the dedicated benchmark_app tool
```
## See Also
- [Integrate the OpenVINO™ Runtime with Your Application](../../../docs/OV_Runtime_UG/integrate_with_your_application.md)
- [Using OpenVINO™ Toolkit Samples](../../../docs/OV_Runtime_UG/Samples_Overview.md)
- [Model Downloader](@ref omz_tools_downloader)
- [Model Optimizer](../../../docs/MO_DG/Deep_Learning_Model_Optimizer_DevGuide.md)
See Also
########
- :doc:`Integrate the OpenVINO™ Runtime with Your Application <openvino_docs_OV_UG_Integrate_OV_with_your_application>`
- :doc:`Using OpenVINO™ Toolkit Samples <openvino_docs_OV_UG_Samples_Overview>`
- :doc:`Model Downloader <omz_tools_downloader>`
- :doc:`Model Optimizer <openvino_docs_MO_DG_Deep_Learning_Model_Optimizer_DevGuide>`
@endsphinxdirective
[openvino.runtime.Core.import_model]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.Core.html#openvino.runtime.Core.import_model
[openvino.runtime.CompiledModel.export_model]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.export_model
[openvino.runtime.Model.add_outputs]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.Model.html#openvino.runtime.Model.add_outputs
[openvino.runtime.set_batch]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.html#openvino.runtime.set_batch
[openvino.runtime.CompiledModel.inputs]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.inputs
[openvino.runtime.CompiledModel.outputs]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.outputs
[openvino.runtime.ConstOutput.any_name]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.ConstOutput.html#openvino.runtime.ConstOutput.any_name
[openvino.runtime.CompiledModel.create_infer_request]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.CompiledModel.html#openvino.runtime.CompiledModel.create_infer_request
[openvino.runtime.InferRequest.infer]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.infer
[openvino.runtime.InferRequest.model_outputs]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.model_outputs
[openvino.runtime.InferRequest.model_inputs]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.model_inputs
[openvino.runtime.InferRequest.query_state]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.query_state
[openvino.runtime.VariableState.reset]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.inference_engine.VariableState.html#openvino.inference_engine.VariableState.reset
[openvino.runtime.InferRequest.profiling_info]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.profiling_info
[openvino.runtime.ProfilingInfo.real_time]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.ProfilingInfo.html#openvino.runtime.ProfilingInfo.real_time
[openvino.runtime.InferRequest.get_input_tensor]:https://docs.openvino.ai/2022.3/api/ie_python_api/_autosummary/openvino.runtime.InferRequest.html#openvino.runtime.InferRequest.get_input_tensor