Skip to content

feat: Add ElevenLabs multilingual TTS/STT evaluation (EN/HI/ES)#113

Open
hopessugar wants to merge 1 commit into
openMF:devfrom
hopessugar:dev
Open

feat: Add ElevenLabs multilingual TTS/STT evaluation (EN/HI/ES)#113
hopessugar wants to merge 1 commit into
openMF:devfrom
hopessugar:dev

Conversation

@hopessugar

@hopessugar hopessugar commented Apr 3, 2026

Copy link
Copy Markdown
Contributor

Summary

Adds ElevenLabs TTS and STT evaluation for multilingual support
as part of the ongoing speech provider benchmarking initiative.

Changes

  • providers/elevenlabs.py — ElevenLabsTTSProvider (Flash v2.5) + ElevenLabsSTTProvider (Scribe v1)
  • tests/test_elevenlabs_multilingual.py — benchmark runner for EN/HI/ES
  • results/elevenlabs/ — JSON results + README with findings
  • requirements.txt — added elevenlabs dependency

Results

Component EN HI ES
TTS Latency 1762ms 647ms 682ms
STT WER 0.0 ✅ 0.089 ✅ 0.0 ✅

Key Findings

  • ElevenLabs Flash v2.5 TTS works well across all 3 languages
  • Scribe v1 STT is excellent - perfect for EN/ES, very good for Hindi
  • Major improvement over Cartesia Hindi STT (WER 1.04 → 0.089)

Latency Measurement

latency_ms = (end_time - start_time) × 1000 (Python time module)

- Add ElevenLabsTTSProvider using Flash v2.5 model
- Add ElevenLabsSTTProvider using Scribe v1 model
- Add multilingual benchmark test for EN/HI/ES
- Add results for all 3 languages (TTS + STT)
- Update requirements.txt with elevenlabs dependency
- TTS fast for HI/ES (~647-682ms), EN slower (1762ms)
- STT perfect for EN/ES (WER 0.0), Hindi very good (WER 0.089)
- Major improvement over Cartesia Hindi STT (1.04 -> 0.089)
@hopessugar hopessugar requested a review from a team April 3, 2026 10:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant