v0.2.9-beta

African Whisper v0.2.9 - beta

This beta release is aimed at identifying potential bugs and areas for improvement.

Release Highlights:

Introduction of mixed precision computation (fp16) to src/training/gradio_inference.py pipeline initialization when CUDA is available in the environment.
Addition of a faster SpeechInference class for transcribing and translating audio files for the API endpoint at src/deployment/main, also featuring mixed precision computation (fp16).

What's Changed

• build(deps): bump yt-dlp from 2023.10.13 to 2023.11.14 by @dependabot in #62
• Introducing SpeechInference Class, CUDA Support, and Dependency Cleanup by @KevKibe in #63

v0.2.8 - beta

African Whisper v0.2.8 - beta

This release being beta is for testing to uncover potential bugs and potential improvement areas.

This release features:

• fix for yt-dlp package conflicts.

Kindly submit an issue if you encounter a bug or potential improvement area

v0.2.7-beta

African Whisper v0.2.7 - beta

This release being beta is for testing to uncover potential bugs and potential improvement areas.

This release features:

• Fix for WandbCallback for predictions on the test dataset.
• Feature for optional training parameters in Trainer Class.

Kindly submit an issue if you encounter a bug or potential improvement area.

v0.2.6 - beta

African Whisper v0.2.6 - beta

This release being beta is for testing to uncover potential bugs and potential improvement areas.

This release features:

• Transcription and translation service of YouTube videos from URL on Gradio UI

Kindly submit an issue if you encounter a bug or potential improvement area.

v0.2.5 - beta

African Whisper v0.2.5 - beta

This release being beta is for testing to uncover potential bugs and potential improvement areas.

This release features:

• Faster training and evaluation times achieved through increased batch sizes.
• Faster mixed-precision training on a GPU supported environment using FP16 numerical format for computation.
• Removal of most unnecessary warnings

Submit an issue if you encounter a bug or potential improvement area.

v0.2.4 - beta

This release addresses a bug related to obtaining transcriptions and translations from microphone input within the Gradio UI.

African Whisper v0.2.3-beta - Unstable Release

This is still a beta version for testing to uncover potential bugs.

This release updates the following features:

• Data processing using streaming method instead of batch.
• Translation and transcription options in Gradio demo UI.

African Whisper v0.2.2-beta - Unstable Release

This is still a beta version for testing inorder to uncover potential bugs.
This release updates the following features:

Data processing using streaming method instead of batch.
Translation and transcription options in Gradio demo UI.

v0.2.1-beta

African Whisper v0.2.1-beta - Unstable Release

This is the first release of African Whisper, a Python library for fast fine-tuning and API endpoint deployment of a Whisper model distribution specifically developed to accelerate Automatic Speech Recognition(ASR) for African Languages.

This release is for testing to uncover potential bugs.

This release includes features such as:

• Full or PEFT optimized fine-tuning of any distribution of Whisper using any dataset from Mozilla's common-voice datasets.
• Evaluation of the training job using Weights and Biases.
• Saving the fine-tuned model to HuggingFace Hub.
• Trying out the fine-tuned model using Gradio UI.
• Deploying a REST API endpoint to get inference from the fine-tuned model.
• Containerizing the endpoint using Docker.