https://github.com/neso613/ASR_TFLite

Collection of ASR models for English TFLite models for faster inference.
https://github.com/neso613/ASR_TFLite

Last synced: about 2 months ago
JSON representation

Collection of ASR models for English TFLite models for faster inference.

• Host: GitHub
• URL: https://github.com/neso613/ASR_TFLite
• Owner: neso613
• License: apache-2.0
• Created: 2021-06-01T09:54:44.000Z (over 3 years ago)
• Default Branch: main
• Last Pushed: 2022-02-21T16:27:27.000Z (over 2 years ago)
• Last Synced: 2024-01-22T21:04:00.634Z (8 months ago)
• Language: Jupyter Notebook
• Homepage:
• Size: 27.6 MB
• Stars: 10
• Watchers: 1
• Forks: 4
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-tensorflow-lite - Inference - asr) | Community | (Models with samples / Speech)

README

        
# ASR_TFLite

This repository provides an Automatic Speech Recognition (ASR) models in TensorFlow Lite (TFLite) for TensorFlow 2.x. These models primarily come from two repositories - [asr](https://www.huylenguyen.com/asr) and [TensorFlowASR](https://github.com/TensorSpeech/TensorFlowASR). We provide end-to-end Jupyter Notebooks that show the inference process using TFLite. \

[English-ASR pip wheel](https://pypi.org/project/english-asr/1.2/)\

[TF Hub](https://tfhub.dev/neso613/lite-model/ASR_TFLite/pre_trained_models/English/1)

## Installation

- tensorflow

- numpy

- librosa

## Models

- [Conformer Transducer](https://arxiv.org/abs/2005.08100) using [LibriSpeech](http://www.openslr.org/12) dataset.

## References

- [Android demo app using the converted model](https://github.com/windmaple/tflite-asr)

- [TensorFlow Lite Conversion](https://www.tensorflow.org/lite/convert)

- [Float16 quantization in TensorFlow Lite](https://www.tensorflow.org/lite/performance/post_training_float16_quant)

- [Dynamic-range quantization in TensorFlow Lite](https://www.tensorflow.org/lite/performance/post_training_quant)