Add voice chat feature with audio storage

Simplify audio post-processing; set pad token; add generation logging

Fix smooth generation: add num_beams=5 for local languages, disable early_stopping, fix lang_norm variable, add generation logging

Fix audio fade-out: remove unnecessary interpolation when speed_factor=1.0, use scipy.signal.resample for better quality, improve normalization

Update main.py

Fix warnings: add attention_mask support, set pad_token, use do_sample with temperature for /speak and /tts-stream

Merge branch 'space-push'

Update main.py

Fix speak-ai: use _transcribe_impl helper to properly handle audio transcription flow

Add 0.8x playback speed to audio output

ASR lazy-load: per-repo cache under /tmp; writable caches; disable lifespan; Farmlingua branding

Reduce storage: move HF caches to /tmp; store model ckpt in /tmp/models; ASR snapshot_download to /tmp with no symlinks

ASR runtime fetch: use snapshot_download in _get_asr with HF_TOKEN; no Docker prefetch required

Rename to Farmlingua Speech System; enforce default voices; ignore speaker_name; update /tts-stream to defaults

ASR+Ask integration, single /speak, scipy WAV, build-time ASR cache, cleanup

Fix: remove unreachable duplicate TTS block in /speak

Fix FileResponse background task; add StreamingResponse cleanup; add sounddevice and system libs

TTS: use scipy.io.wavfile for WAV output; remove torchcodec dependency path

TTS: robust ckpt handling; download to /tmp then move; make /code writable

Update main.py

push

push

push

push

push