SteerTTS: open-source 23M-param TTS with 12 numeric voice controls, real-time on CPU
Working_Hat5120 · reddit · 2026-10-19
The author released SteerTTS, a small English TTS model you steer with numbers, now open on Hugging Face (weights CC BY-NC 4.0, code MIT).
Architecture & capabilities:
- 9.6M-param acoustic model (attention + bidirectional Mamba-3, FastPitch-style) + Vocos vocoder = 23M total
- 12 controls as z-scores relative to a voice's neutral speech: rate, loudness, median pitch, pitch range, final rise/fall, pauses, spectral tilt, breathiness measures, etc.
- 51 style tags, free-text prompts (resolved into inspectable JSON), inline markup (emphasis, [pause 400ms]), sentence-level streaming
- 2,825 built-in anonymized voices plus experimental zero-shot cloning
Performance: RTFx 60–80 on an RTX 5070; 4–8 on CPU without the optional flow refiner, first streamed audio in 0.5s; 1.8% average WER on Open TTS Leaderboard English sets (local run, unofficial).
Honest limitations: loudness/pitch/tilt steer reliably but pause, voicing and CPPS controls are weak; emotion tags act as style switches a classifier can't distinguish; cloning similarity is low; free-text prompts struggle with negation; English only, no human evals yet.
More from Multimodal
- Full AI virtual influencer character sheet: the complete template for building a creator persona — CurieuxExplorer · 2026-10-19
- Midjourney v6.1 still delivers: sref + raw parameter recipe for bold comic-style portraits — michaelrabone · 2026-10-19
- 12-task edit benchmark: smallest FLUX.3 Klein lands just 1 point behind Qwen Image 2.1 — Diligent-Rub-2113 · 2026-10-19
- How Mission Pandora keeps viewers watching a 35-minute AI film: log-entry storytelling — lmoroney · 2026-10-19
- Consistent Distribution Matching: data-free diffusion distillation beats DMD2 with 20GB VRAM — vector-institute · 2026-10-19
- Topview open-sources 3D Builder for blocking shots before AI video generation — Aiden_Tech_Ai · 2026-10-19