compare
Speaches vs WhisperX
The same facts for both, read from GitHub every night, and the relation a person reviewed.
| Fact | Speaches | WhisperX |
|---|---|---|
| Language | Python | Python |
| Licence | MIT | BSD-2-Clause |
| Stars | 3.7k | 25k |
| Latest | v0.9.0-rc.3 | v3.8.6 |
| Last push | 2026-10-17 | 2026-10-18 |
| Release cadence | about 46 days between releases | about 7 days between releases |
| Active contributors | 0 commit authors on the default branch in the last 90 days | 3 commit authors on the default branch in the last 90 days |
| Flags | none | none |
How they relate
Both replace Amazon Transcribe. Alternatives to Amazon Transcribe →
PartialSpeachesOpenAI API-compatible server for streaming transcription with faster-whisper, run on your own hardware.
PartialWhisperXBatch transcription with speaker labels on your own hardware; no managed service or streaming.
Speaches
- v0.9.0-rc.32025-12-27tests: add additional speech endpoint tests
- v0.9.0-rc.22025-11-11lots of changes
- v0.9.0-rc.12025-09-25chore(deps): update pre-commit hook pre-commit/pre-commit-hooks to v6 by @renovate[bot] inpre-release
- v0.8.32025-09-19fix: corrupted headers for wav and flac audio formats when streaming
- v0.8.22025-09-19deps: update default cuda version
WhisperX
- 3.8.7rc12026-06-26Fix Windows CUDA detection: include AMD64 in uv source platform markerspre-release
- v3.8.62026-05-25Add Indonesian model to alignment.py
- v3.8.52026-04-01fix: pin torchvision and torchcodec for torch 2.8 compatibility
- v3.7.92026-03-25Backport of word-level timestamp fixes from v3.8.4.
- v3.5.22026-03-25Backport of word-level timestamp fixes from v3.8.4.
Speaches
unsigned The latest release, v0.9.0-rc.3, carries no signature GitHub could verify.
Loading the security report
WhisperX
unsigned The latest release, v3.8.6, carries no signature GitHub could verify.
Loading the security report