deepasr 0.0.3

Keras(Tensorflow) implementations of Automatic Speech Recognition

DeepAsr

DeepAsr is an open-source implementation of end-to-end Automatic Speech Recognition (ASR) engine.

DeepAsr provides multiple Speech Recognition architectures, Currently it provides Baidu's Deep Speech 2 using Keras (Tensorflow).

Using DeepAsr you can:

• perform speech-to-text using pre-trained models
• tune pre-trained models to your needs
• create new models on your own

DeepAsr key features:

• Multi GPU support: You can do much more like distribute the training using the Strategy, or experiment with mixed precision policy.
• CuDNN support: Model using CuDNNLSTM implementation by NVIDIA Developers. CPU devices is also supported.
• DataGenerator: The feature extraction (on CPU) can be parallel to model training (on GPU).

Installation

You can use pip:

pip install deepasr

Getting started

The speech recognition is a tough task. You don't need to know all details to use one of the pretrained models. However it's worth to understand conceptional crucial components:

• Input: WAVE files with mono 16-bit 16 kHz (up to 5 seconds)
• FeaturesExtractor: Convert audio files using MFCC Features or Spectrogram
• Model: CTC model defined in Keras (references: [1], [2])
• Decoder: Greedy algorithm with the language model support decode a sequence of probabilities using Alphabet
• DataGenerator: Stream data to the model via generator
• Callbacks: Set of functions monitoring the training

import numpy as np
import pandas as pd
import tensorflow as tf
import deepasr as asr

def get_config(features, multi_gpu):
    alphabet_en = asr.vocab.Alphabet(lang='en')

    features_extractor = asr.features.preprocess(feature_type=features, features_num=161,
                                                 samplerate=16000,
                                                 winlen=0.02,
                                                 winstep=0.01,
                                                 winfunc=np.hanning)

    model = asr.model.get_deepspeech2(
        input_dim=161,
        output_dim=29,
        is_mixed_precision=True
    )
    optimizer = tf.keras.optimizers.Adam(
        lr=1e-4,
        beta_1=0.9,
        beta_2=0.999,
        epsilon=1e-8
    )
    decoder = asr.decoder.GreedyDecoder()

    pipeline = asr.pipeline.ctc_pipeline.CTCPipeline(
        alphabet=alphabet_en, features_extractor=features_extractor, model=model, optimizer=optimizer, decoder=decoder,
        sample_rate=16000, mono=True, multi_gpu=multi_gpu
    )
    return pipeline


def run(train_data, test_data, features='fbank', batch_size=32, epochs=10, multi_gpu=True):
    pipeline = get_config(features, multi_gpu)
    # history = pipeline.fit_iter(train_data, batch_size=batch_size, epochs=epochs, iter_num=1000)
    history = pipeline.fit_generator(train_data, batch_size=batch_size, epochs=epochs)
    pipeline.save('./checkpoints')
    print("Truth:", test_data['transcript'].to_list()[0])
    print("Prediction", pipeline.predict(test_data['path'].to_list()[0]))
    return history


train = pd.read_csv('train_data.csv')
test = pd.read_csv('test_data.csv')
run(train, test, features='fbank', batch_size=32, epochs=100, multi_gpu=True)

Loaded pre-trained model has all components. The prediction can be invoked just by calling pipline.predict().

import pandas as pd
import deepasr as asr
pipeline = asr.pipeline.load('./checkpoints')
test_data = pd.read_csv('test_data.csv')
print("Truth:", test_data['transcripts'].to_list()[0])
print("Prediction", pipeline.predict(test_data['path'].to_list()[0]))

References

The fundamental repositories:

• Baidu - DeepSpeech2 - A PaddlePaddle implementation of DeepSpeech2 architecture for ASR
• NVIDIA - Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP
• TensorFlow - The implementation of DeepSpeech2 model
• Mozilla - DeepSpeech - A TensorFlow implementation of Baidu's DeepSpeech architecture
• Espnet - End-to-End Speech Processing Toolkit
• Automatic Speech Recognition - Distill the Automatic Speech Recognition research