fix(stt): provider robustness — device config, CUDA fallbacks, container normalization (SILK/m4a), lazy-install recovery - #73510

Merged

teknium1 merged 23 commits into

mainfrom

fix/voice-stt-provider-robustness

Jul 28, 2026

Commits on Jul 28, 2026

fix(stt): respect device and compute_type from config.yaml

Tranquil-Flow
authored and
teknium1
committed
1fdd8ba
fix(stt): treat CUBLAS_STATUS_NOT_SUPPORTED as CUDA lib error

liuhao1024
authored and
teknium1
committed
5d86ba6
fix: avoid local STT crash on Apple Silicon

AnthonyAssistantAi
authored and
teknium1
committed
abd5bdf
fix(stt): scope upload size limits to remote providers

ypwcharles
authored and
teknium1
committed
4d89d79
fix(tts): fall through to raw import when lazy_deps fails (#53259)

tusharui
authored and
teknium1
committed
027a2dd
test(tts): add STT fallback regression for _transcribe_mistral (#53259)

tusharui
authored and
teknium1
committed
99d66d6
fix: handle missing transcription module gracefully

RichardHojunJang
authored and
teknium1
committed
4ebad50
fix(stt): report unregistered configured providers

ooiuuii
authored and
teknium1
committed
3162cb5
fix(stt): strip Qwen3-ASR response prefix

LauraGPT
authored and
teknium1
committed
bc036cb
fix(stt): anchor Qwen3-ASR envelope stripping

LauraGPT
authored and
teknium1
committed
113b67f
fix(stt): anchor qwen asr envelope stripping

LauraGPT
authored and
teknium1
committed
fbfe93d
fix(stt): check_voice_requirements() should recognize all STT providers

zehuaw1
authored and
teknium1
committed
1a1445a
fix(stt): check selected provider (not any) + plugin support

zehuaw1
authored and
teknium1
committed
ad38396
fix(stt): validate selected voice provider availability

zehuaw1
authored and
teknium1
committed
ba81419
fix(stt): better error logging when faster-whisper lazy install fails

damiankluk
authored and
teknium1
committed
b2975f5
fix(stt): preprocess .silk voice notes before transcription

dso2ng
authored and
teknium1
committed
617f770
transcription: transcode to m4a and retry when OpenAI STT rejects the audio container

carljborg
authored and
teknium1
committed
4331d44
fix(stt): lock local model load; allow keyless local OpenAI-compatible STT

teknium1
and
nnnet
committed
1fc603c
chore: add contributor email mappings for salvaged STT commits
teknium1
committed
d8d4d75
test: align dispatch tests with provider-scoped validation and named registration errors
teknium1
committed
222662d
test: add BadRequestError to the fake openai module fixture
teknium1
committed
213fbd2
fix: explicit utf-8 encoding on ffmpeg STT transcode subprocess (Windows footgun lint)
teknium1
committed
b569b6b
fix: stdin=DEVNULL + windows_hide_flags on STT transcode subprocess (guard test)
teknium1
committed
fa5bf44

Navigation Menu

fix(stt): provider robustness — device config, CUDA fallbacks, container normalization (SILK/m4a), lazy-install recovery - #73510

fix(stt): provider robustness — device config, CUDA fallbacks, container normalization (SILK/m4a), lazy-install recovery#73510
teknium1 merged 23 commits into
mainfrom
fix/voice-stt-provider-robustness

Commits on Jul 28, 2026

fix(stt): respect device and compute_type from config.yaml

fix(stt): treat CUBLAS_STATUS_NOT_SUPPORTED as CUDA lib error

fix: avoid local STT crash on Apple Silicon

fix(stt): scope upload size limits to remote providers

fix(tts): fall through to raw import when lazy_deps fails (#53259)

test(tts): add STT fallback regression for _transcribe_mistral (#53259)

fix: handle missing transcription module gracefully

fix(stt): report unregistered configured providers

fix(stt): strip Qwen3-ASR response prefix

fix(stt): anchor Qwen3-ASR envelope stripping

fix(stt): anchor qwen asr envelope stripping

fix(stt): check_voice_requirements() should recognize all STT providers

fix(stt): check selected provider (not any) + plugin support

fix(stt): validate selected voice provider availability

fix(stt): better error logging when faster-whisper lazy install fails

fix(stt): preprocess .silk voice notes before transcription

transcription: transcode to m4a and retry when OpenAI STT rejects the audio container

fix(stt): lock local model load; allow keyless local OpenAI-compatible STT

chore: add contributor email mappings for salvaged STT commits

test: align dispatch tests with provider-scoped validation and named registration errors

test: add BadRequestError to the fake openai module fixture

fix: explicit utf-8 encoding on ffmpeg STT transcode subprocess (Windows footgun lint)

fix: stdin=DEVNULL + windows_hide_flags on STT transcode subprocess (guard test)