TensorSpeech/TensorFlowASR

:zap: TensorFlowASR: Almost State-of-the-art Automatic Speech Recognition in Tensorflow 2. Supported languages that can use characters or subwords

What it solves

It provides a comprehensive framework for building and deploying Automatic Speech Recognition (ASR) systems, enabling the conversion of spoken audio into text using various state-of-the-art neural network architectures.

How it works

The project implements multiple ASR architectures categorized by their training loss functions: Transducer models (like Conformer and ContextNet) using RNNT loss, and CTC models (like DeepSpeech2 and Jasper) using CTC loss. It includes a full pipeline covering feature extraction, data augmentation, training, and decoding (greedy and beam search). For deployment, models can be converted to TFLite to reduce memory and computation requirements.

Who it’s for

Developers and researchers working on speech-to-text applications who prefer the TensorFlow ecosystem and need a flexible toolkit for training, testing, and deploying ASR models on various hardware, including CPUs, NVIDIA GPUs, and Cloud TPUs.

Highlights

  • Supports multiple architectures including Conformer, ContextNet, RNN Transducer, DeepSpeech2, and Jasper.
  • Provides TFLite conversion for efficient deployment.
  • Includes support for both streaming and non-streaming inference.
  • Features built-in support for common ASR datasets in English and Vietnamese.

Related

  • Project
    tensorforce/tensorforce

    A modular TensorFlow-based deep reinforcement learning framework for research and practice, providing flexible components to build and deploy RL agents.

  • Project
    inworld-ai/tts

    A training and modeling framework for SpeechLM-based text-to-speech systems, supporting pre-training, SFT, and RLHF alignment across single or multi-GPU clusters.

  • Project
    LongxingTan/Time-series-prediction

    TFTS is a TensorFlow-based library that provides a unified interface for implementing state-of-the-art deep learning models for time series prediction, classification, and anomaly detection.

  • Project
    lenML/Speech-AI-Forge

    A unified framework and API server providing a centralized interface for multiple TTS and ASR models, including ChatTTS, CosyVoice and Whisper.

  • Project
    QwenAudio/SenseVoice

    SenseVoice is a multilingual speech foundation model that provides high-accuracy speech recognition, emotion detection, and audio event identification with low-latency inference.