Update STT model to large-v3-turbo#267
Conversation
📝 WalkthroughWalkthroughA configuration value in the FasterWhisper STT plugin was updated, changing the model identifier from Changes
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~2 minutes Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@ovos_config/recommends/gpu/en-us.conf`:
- Line 6: The model identifier value for the "model" key is inconsistent:
en-us.conf uses "large-v3-turbo" while other GPU language configs use
"whisper-large-v3-turbo"; decide which canonical identifier to use (either the
bare CTranslate2 target "large-v3-turbo" or the explicit
"whisper-large-v3-turbo"), then update the divergent files so they all use the
same string; specifically locate the "model" entries in the per-language GPU
config files (e.g., en-us.conf, nl-nl.conf, it-it.conf, de-de.conf, fr-fr.conf,
da-dk.conf) and replace the non-canonical value so all files match, and run any
config validation or integration tests to ensure FasterWhisper/CTranslate2
loading still works with the chosen identifier.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 3d3c114b-8322-4a2e-85ff-fabd495c29d7
📒 Files selected for processing (1)
ovos_config/recommends/gpu/en-us.conf
| "fallback_module": "", | ||
| "ovos-stt-plugin-fasterwhisper": { | ||
| "model": "whisper-large-v3-turbo", | ||
| "model": "large-v3-turbo", |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Description: Search for FasterWhisper model configurations and verify consistency
# Search for all FasterWhisper model configurations
echo "=== Searching for ovos-stt-plugin-fasterwhisper model configurations ==="
rg -n --type=json -A2 -B2 'ovos-stt-plugin-fasterwhisper'
echo -e "\n=== Searching for whisper model references ==="
rg -n 'whisper.*turbo|turbo.*whisper|large-v3-turbo' --glob='*.conf' --glob='*.json'
echo -e "\n=== Checking other GPU/STT config files ==="
fd -e conf -e json . ovos_config/recommends --exec echo "File: {}" \; --exec cat {}Repository: OpenVoiceOS/ovos-config
Length of output: 41075
🌐 Web query:
ovos-stt-plugin-fasterwhisper model identifier format large-v3-turbo
💡 Result:
The model identifier format for ovos-stt-plugin-fasterwhisper is a simple string name like "large-v3" for standard Whisper models, or a full Hugging Face repository ID/path for custom models (e.g., "projecte-aina/faster-whisper-large-v3-ca-3catparla"). Supported standard model identifiers include: 'tiny.en', 'tiny', 'base.en', 'base', 'small.en', 'small', 'medium.en', 'medium', 'large-v1', 'large-v2', 'large-v3', 'large', 'distil-large-v2', 'distil-medium.en', 'distil-small.en', 'distil-large-v3'. "large-v3-turbo" is not listed among the standard identifiers in the official plugin documentation. However, since the plugin uses faster-whisper (which supports CTranslate2-converted Whisper models), "large-v3-turbo" can be used directly if referring to a compatible converted model. Community configurations explicitly use "whisper-large-v3-turbo" or "large-v3-turbo" as the model string, and faster-whisper loads models like "dropbox-dash/faster-whisper-large-v3-turbo" (the CTranslate2 conversion of openai/whisper-large-v3-turbo). Thus, the format allows "large-v3-turbo" as a valid identifier, likely mapping to the turbo variant.
Citations:
- 1: https://github.com/OpenVoiceOS/ovos-stt-plugin-fasterwhisper
- 2: https://blog.graywind.org/posts/fasterwhisper-stt-server-script/
- 3: https://github.com/OpenVoiceOS/ovos-docker-stt
- 4: https://www.huggingface.co/openai/whisper-large-v3-turbo
- 5: https://huggingface.co/dropbox-dash/faster-whisper-large-v3-turbo
Inconsistency: en-us.conf differs from all other language variants in model identifier format.
The model identifier changed from whisper-large-v3-turbo to large-v3-turbo, but five other GPU language configs (nl-nl, it-it, de-de, fr-fr, da-dk) still use whisper-large-v3-turbo. While the web search confirms both formats may work with FasterWhisper's support for CTranslate2-converted models, this cross-file inconsistency suggests either an incomplete migration across all language files or a language-specific fix for issue #266. Verify that either all GPU language configs should be updated to large-v3-turbo, or confirm that en-us requires a distinct format.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@ovos_config/recommends/gpu/en-us.conf` at line 6, The model identifier value
for the "model" key is inconsistent: en-us.conf uses "large-v3-turbo" while
other GPU language configs use "whisper-large-v3-turbo"; decide which canonical
identifier to use (either the bare CTranslate2 target "large-v3-turbo" or the
explicit "whisper-large-v3-turbo"), then update the divergent files so they all
use the same string; specifically locate the "model" entries in the per-language
GPU config files (e.g., en-us.conf, nl-nl.conf, it-it.conf, de-de.conf,
fr-fr.conf, da-dk.conf) and replace the non-canonical value so all files match,
and run any config validation or integration tests to ensure
FasterWhisper/CTranslate2 loading still works with the chosen identifier.
1 participant
This solve #266
Summary by CodeRabbit