https://github.com/a-iceberg/whisper-timestamped

Timestamped ASR microservice
https://github.com/a-iceberg/whisper-timestamped

asr audio-to-text automatic-speech-recognition data-analysis data-science deep-learning docker fastapi mlops monitoring mssqlserver openai prompt-engineering python resource-management timestamps uvicorn-gunicorn whisper

Last synced: about 1 month ago
JSON representation

Timestamped ASR microservice

• Host: GitHub
• URL: https://github.com/a-iceberg/whisper-timestamped
• Owner: a-iceberg
• Created: 2024-01-24T10:54:11.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2024-11-07T09:37:25.000Z (about 2 months ago)
• Last Synced: 2024-11-07T10:33:47.853Z (about 2 months ago)
• Topics: asr, audio-to-text, automatic-speech-recognition, data-analysis, data-science, deep-learning, docker, fastapi, mlops, monitoring, mssqlserver, openai, prompt-engineering, python, resource-management, timestamps, uvicorn-gunicorn, whisper
• Language: Jupyter Notebook
• Homepage:
• Size: 3.28 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
## Whisper timestamped

ASR microservice developed for [call center transcription service](https://github.com/a-iceberg/stt_server)

Based on [whisper-timestamped](https://github.com/linto-ai/whisper-timestamped)

---

### Results

The latest version of the Whisper model - [v3](https://github.com/openai/whisper/discussions/1762), is used; service can operate on both GPU and CPU, but significantly slower on the latter. [Prompt engineering](https://github.com/a-iceberg/whisper-timestamped/blob/9cb99bddf801b01ee3c187d0909035f8dcaf4aa8/transcribe.py#L70) was applied to improve the transcription results.

Transcription quality on Russian ([source](https://github.com/a-iceberg/whisper_model_evaluator/blob/whisper/reports/whisper_comparator.ipynb)):

* *WER* - **0.2**

* *MER* - **0.2**

* *WIL* - **0.25**