Silencio-2: open 18M-param TTS with emotion tags, ~20x realtime on CPU
N1kuson · reddit · 2026-11-02
A developer released Silencio-2 (BlueRoseLabs, weights CC-BY-4.0, inference code MIT), an open TTS fine-tuned from Kyutai's Pocket TTS 18M that follows inline tags:
- Emotions: happy, sad, angry, fear, surprised, whisper
- Sounds: laugh, sigh, gasp, cough, clearthroat
Key specs: 18.2M params (27.9M with voice-cloning encoder), 112 MB on disk; 20x realtime on an Apple M4 with 4 CPU threads, no GPU; 24 kHz, streaming, voice cloning from a short clip; trained on 2x T4 in 6 hours on 50 hours of audio.
Training postmortem (the most useful part):
- Run 1: tags trained but did almost nothing
- Run 2: filtering data with an AudioSet tagger silently removed all laugh/cough clips — discovered only by counting rows per tag
- Run 3: switched to 3s sliding windows; 0 of 23 scripted-laughter clips passed threshold; run crashed on T4
- Run 4: removed the filter entirely. Lesson: clip-level audio taggers are poor gates for short sounds in speech; count rows per tag and spot-check by ear
Weaknesses: whisper is the least stable tag; tags vary by seed; no formal WER run (expect 1.9% vs 1.5% for base).
More from Multimodal
- Midjourney textures get a retro Dreamcast makeover via custom workflow — AIandDesign · 2026-11-02
- Video gen on AMD 9070 XT: 5 seconds of 480p takes 8 minutes with MiniMax H3 — darthrehad · 2026-11-02
- Redditor shares cinematic AI-generated original music video — Hellish_NDE · 2026-11-02
- I had a dream, then I generated it: the Backrooms as a waterslide — IamIfOnly · 2026-11-02
- Moonlit museum video demo made with Seedance 2.5 on Dreamina — LudovicCreator · 2026-11-02
- ComfyUI Plugin Adds 1400+ Visual Style Picker for Krea 2 and Prompt Preservation — skk80 · 2026-11-02