ggml-org/whisper.cpp

tests : add WER benchmarks

Open

#2,454 opened on Oct 5, 2024

View on GitHub
聽(26 comments)聽(0 reactions)聽(0 assignees)C++聽(49,693 stars)聽(5,535 forks)batch import
help wantedhigh priorityresearch馃敩roadmap

Description

It would be nice to start measuring the word error rate (WER) of whisper.cpp across some representative dataset:

  • short audio
  • long audio
  • english
  • non-english
  • etc.

This will help us catch regressions in the future. I'm not familiar with what is typically used for TTS WER benchmarks, so looking for help from the community.

Contributor guide

Tech stack
ccpp
Domain
testingmachine learningperformance
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
3-5 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
Basic understanding of whisper.cppFamiliarity with word error rate (WER)Experience with audio datasets
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
The issue requests adding WER benchmarks to whisper.cpp. Start by researching standard ASR benchmark datasets (e.g., LibriSpeech, Common Voice) that include short/long audio and English/non English samples. Examine the existing test infrastructure in the repository (likely in tests/ directory or Makefile) to understand how to integrate new benchmarks. Review the comments on the issue for community suggestions on dataset selection and evaluation methodology. Coordinate with maintainers to agree on a concrete set of datasets and metrics before implementing.
tests : add WER benchmarks 路 ggml-org/whisper.cpp#2454 | Good First Issue