Indic-Mio

This first-party Soniqo page documents Indic-Mio from the local speech-swift / speech-core implementation. Hugging Face bundles are linked below after the integration notes.

Internal Page First

Landing cards and docs menus now point here first; source model and bundle links remain available from this page.

At a Glance

Model	Indic-Mio
Role	Hindi / Indic emotion TTS with optional reference voice conditioning
Backend	MLX fp16
Output	24 kHz mono waveform
Languages	Hindi / Indic text focus
License	Bundle-specific; verify upstream terms before commercial use
Status	Programmatic SpeechGenerationModel runtime
Source	Indic-Mio bundle
Swift product	`IndicMioTTS`
CLI / runtime	Programmatic runtime; server-side loader support is present

Use

The snippet below mirrors the current speech-swift API or command exposed by the repo.

import IndicMioTTS

let model = try await IndicMioTTSModel.fromPretrained()
let pcm = try await model.generate(
    text: "नमस्ते <happy>",
    language: "hi"
)

Model Links

Implementation Notes

Download requests the primary model, tokenizer/chat files, and miocodec component explicitly; the optional WavLM reference encoder is fetched separately when cloning is used.
Supported emotion markers are <happy>, <sad>, <angry>, <disgust>, <fear>, and <surprise>.
MioCodec uses 25 Hz content tokens and decodes to 24 kHz audio.
Reference cloning extracts a 128-float global embedding through the WavLM reference encoder path.