Skip to content
ModelListedExternal
Browse files

Chatterbox Flash CoreML

Compiled Core ML export of ResembleAI/chatterbox-flash for Apple runtimes.

Sourcehuggingface.co/aufklarer/Chatterbox-Flash-CoreML/tree/e845c9c0f1a5ac50a6f1f900a40006acfe640e58
KindOtherBase modelResembleAI/chatterbox-flashVersionRevision e845c9cCgrade

Model source

Kind
Other
File size
243 B–1,000 MB
Base model
ResembleAI/chatterbox-flash
Version
vRevision e845c9c
Updated
Oct 31, 2026
Downloads
77
Source
Hugging Face

Source description

Chatterbox Flash CoreML

Compiled Core ML export of ResembleAI/chatterbox-flash for Apple runtimes.

This bundle contains the Chatterbox Flash T3 token generator and the S3Gen audio back-half as static-shape .mlmodelc graphs. It is intended for an application runtime that owns text tokenization, T3 denoising/sampling, reference conditioning, and graph orchestration.

Links

Model

ComponentParametersFormatPrecisionStatic shape
T3 block-diffusion token generator532.4MCore ML .mlmodelcfp16text_len 256, block_size 16, max_seq 1024
S3Gen audio back-half266.0MCore ML .mlmodelcfp16token_len 192, mel_len 384
Total798.4MCore ML .mlmodelcfp1624 kHz waveform output

Files

PathSizeDescription
config.json4 KBRoot metadata for download tracking and runtime discovery
t3/ConditioningEncoder.mlmodelc25 MBSpeaker/prompt/emotion conditioning to T3 conditioning embedding
t3/TextPrefill.mlmodelc963 MBCausal [cond, text, start_speech] prefix prefill and flat KV cache
t3/BlockDecoder.mlmodelc1.0 GBFull-visible Flash speech-block logits with explicit KV cache
t3/uncond_block_prior.npy36 KBUnconditional PMI prior for Flash scoring
t3/tokenizer.json28 KBChatterbox Flash text tokenizer
t3/config.json4 KBT3 export metadata
audio/FlowSpeakerProjector.mlmodelc44 KBS3Gen reference embedding to projected speaker conditioning
audio/FlowEncoder.mlmodelc79 MBprompt_token ++ speech_tokens to flow mu and mask
audio/FlowEstimator.mlmodelc141 MBOne meanflow Euler derivative step
audio/HiFTVocoder.mlmodelc41 MBMel frames to 24 kHz waveform
audio/audio_config.json4 KBS3Gen audio export metadata

Runtime Boundary

The exported graphs cover:

  • T3 conditioning, text prefill, and block decoding.
  • S3Gen speaker projection, flow encoder, meanflow estimator step, and HiFT vocoder.

The host runtime must still provide:

  • text normalization and tokenization
  • T3 denoising loop, PMI/CFG scoring, sampling, and EOS trimming
  • reference waveform encoders:
    • VoiceEncoder: ref.wav -> speaker_emb
    • prompt speech tokenizer: ref.wav -> prompt_speech_tokens
    • S3Gen reference encoder: ref.wav -> prompt_token, prompt_feat, embedding
  • padding/cropping and the two-step meanflow loop over (t,r) = (0,0.5), (0.5,1.0)

This means the bundle supports voice-cloning TTS when the runtime supplies reference conditioning tensors, but it is not yet a fully Core ML ref.wav -> cloned wav pipeline.

Validation

TestResult
T3 graph roundtrip vs PyTorch wrappersPass at 2% relative tolerance
Audio graph roundtrip vs PyTorch wrappersPass for token_len 192, mel_len 384
Stitched S3Gen meanflow-to-audio roundtripPass
Prompted synthesis smoke testPass
Whisper tiny transcript of generated wavCore ML speech test.
Smoke-test WER0.000

Core ML warnings from local export:

  • CPU_ONLY prediction crashed for the T3 package in local coremltools 8.3 testing. Use ALL, CPU_AND_NE, or a compiled-device runtime.
  • The uploaded artifact ships compiled .mlmodelc folders. The numerical parity tests were run against the source .mlpackage exports before compilation.

Usage Sketch

The runtime loads graphs from t3/ and audio/, then:

  1. Prepare T3 reference conditioning tensors and S3Gen ref_dict tensors from a prompt wav.
  2. Tokenize text with t3/tokenizer.json.
  3. Run T3 prefill and block denoising/sampling to produce S3 speech tokens.
  4. Concatenate S3Gen prompt_token and generated speech tokens.
  5. Run audio/FlowEncoder.mlmodelc.
  6. Build cond by copying prompt_feat.T into the mel prefix.
  7. Run audio/FlowEstimator.mlmodelc twice for (0,0.5) and (0.5,1.0).
  8. Crop generated mel frames after the prompt prefix.
  9. Run audio/HiFTVocoder.mlmodelc and crop padded samples.

Source

Converted from ResembleAI/chatterbox-flash, revision 4385507288b8197e6dab8b4e6b1603328d549d9d.

Sources

1 source
Hugging FacePrimaryVersion e845c9c0f1a5ac50a6f1f900a40006acfe640e58
huggingface.co/aufklarer/Chatterbox-Flash-CoreML/tree/e845c9c0f1a5ac50a6f1f900a40006acfe640e58

Verified Oct 31

Download files

Repository files

File inventory only; completeness is unverified.

Model artifacts

21 artifacts

audio/FlowEncoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 0f2d472977d4…f2b1 · Hugging Face

Download file

audio/FlowEncoder.mlmodelc/coremldata.bin

bin · 358 B · SHA-256 0577fe1766ab…8892 · Hugging Face

Download file

audio/FlowEncoder.mlmodelc/weights/weight.bin

bin · 78.9 MB · SHA-256 777756b37ea8…b456 · Hugging Face

Download file

audio/FlowEstimator.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 a0cf95065b64…63b0 · Hugging Face

Download file

audio/FlowEstimator.mlmodelc/coremldata.bin

bin · 423 B · SHA-256 0dad9d8ceb14…f604 · Hugging Face

Download file

audio/FlowEstimator.mlmodelc/weights/weight.bin

bin · 140 MB · SHA-256 21aa1d17d6ff…fd20 · Hugging Face

Download file

audio/FlowSpeakerProjector.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 b256dfdf75f8…88b9 · Hugging Face

Download file

audio/FlowSpeakerProjector.mlmodelc/coremldata.bin

bin · 312 B · SHA-256 7d5ce09013a5…f2e4 · Hugging Face

Download file

audio/FlowSpeakerProjector.mlmodelc/weights/weight.bin

bin · 30.3 KB · SHA-256 862da38b6b14…e929 · Hugging Face

Download file

audio/HiFTVocoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 b0e902f149c0…0059 · Hugging Face

Download file

audio/HiFTVocoder.mlmodelc/coremldata.bin

bin · 310 B · SHA-256 746fbf67e3bc…3fbe · Hugging Face

Download file

audio/HiFTVocoder.mlmodelc/weights/weight.bin

bin · 41.1 MB · SHA-256 84ccb67963f5…eebd · Hugging Face

Download file

t3/BlockDecoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 1497bad8a0b3…7cd8 · Hugging Face

Download file

t3/BlockDecoder.mlmodelc/coremldata.bin

bin · 606 B · SHA-256 75669cb63bde…218e · Hugging Face

Download file

t3/BlockDecoder.mlmodelc/weights/weight.bin

bin · 1,000 MB · SHA-256 f9852f0b19bd…af2f · Hugging Face

Download file

t3/ConditioningEncoder.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 4f5dcb86707d…be9d · Hugging Face

Download file

t3/ConditioningEncoder.mlmodelc/coremldata.bin

bin · 385 B · SHA-256 340d00728a03…fe4d · Hugging Face

Download file

t3/ConditioningEncoder.mlmodelc/weights/weight.bin

bin · 24.9 MB · SHA-256 c89c6626ea57…fff1 · Hugging Face

Download file

t3/TextPrefill.mlmodelc/analytics/coremldata.bin

bin · 243 B · SHA-256 a3e6abc5a11c…909c · Hugging Face

Download file

t3/TextPrefill.mlmodelc/coremldata.bin

bin · 412 B · SHA-256 0ea140528b70…a03e · Hugging Face

Download file

t3/TextPrefill.mlmodelc/weights/weight.bin

bin · 962 MB · SHA-256 48c7c8fc9152…f2e4 · Hugging Face

Download file

Source excerpts

3 excerpts
ReadmeAug 15huggingface.co/aufklarer/Chatterbox-Flash-CoreML

--- library_name: coremltools base_model: - ResembleAI/chatterbox-flash language: - en license: mit tags: - coreml - text-to-speech - voice-cloning - zero-shot-tts - block-diffusion - speech-synthesis - apple-silicon - ios - macos pipeline_tag: text-to-speech --- # Chatterbox Flash CoreML Compiled Core ML export of [ResembleAI/chatterbox-flash](https://huggingface.co/ResembleAI/chatterbox-flash) for Apple runtimes. This bundle contains the Chatterbox Flash T3 token generator and the S3Gen audio back-half as static-shape `.mlmodelc` graphs. It is intended for an application runtime that owns text tokenization, T3 denoising/sampling, reference conditioning, and graph orchestration. ## Links - [speech-swift](https://github.com/soniqo/speech-swift) — Apple SDK - [Speech Studio](https://soniqo.audio/speech-studio) — local speech generation and voice cloning app - [Docs](https://soniqo.audio/getting-started) — install and CLI docs - [soniqo.audio](https://soniqo.audio) — website - [blog](https://soniqo.audio/blog) — blog ## Model | Component | Parameters | Format | Precision | Static shape | |---|---:|---|---|---| | T3 block-diffusion token generator | 532.4M | Core ML `.mlmodelc` | fp16 | text_len 256, block_size 16, max_seq 1024 | | S3Gen audio back-half | 266.0M | Core ML `.mlmodelc` | fp16 | token_len 192, mel_len 384 | | Total | 798.4M | Core ML `.mlmodelc` | fp16 | 24 kHz waveform output | ## Files | Path | Size | Description | |---|---:|---| | `config.json` | 4 KB | Root metadata for download tracking and runtime discovery | | `t3/ConditioningEncoder.mlmodelc` | 25 MB | Speaker/prompt/emotion conditioning to T3 conditioning embedding | | `t3/TextPrefill.mlmodelc` | 963 MB | Causal `[cond, text, start_speech]` prefix prefill and flat KV cache | | `t3/BlockDecoder.mlmodelc` | 1.0 GB | Full-visible Flash speech-block logits with explicit KV cache | | `t3/uncond_block_prior.npy` | 36 KB | Unconditional PMI prior for Flash scoring | | `t3/tokenizer.json` | 28 KB | Chatterbox Flash text tokenizer | | `t3/config.json` | 4 KB | T3 export metadata | | `audio/FlowSpeakerProjector.mlmodelc` | 44 KB | S3Gen reference embedding to projected speaker conditioning | | `audio/FlowEncoder.mlmodelc` | 79 MB | `prompt_token ++ speech_tokens` to flow `mu` and mask | | `audio/FlowEstimator.mlmodelc` | 141 MB | One meanflow Euler derivative step | | `audio/HiFTVocoder...

NoteOct 31huggingface.co/aufklarer/Chatterbox-Flash-CoreML

Source context: 77 downloads · 0 likes · Pipeline text-to-speech · Library coremltools · Repo aufklarer/Chatterbox-Flash-CoreML

NoteAug 15huggingface.co/aufklarer/Chatterbox-Flash-CoreML

Source context: 57 downloads · 0 likes · Pipeline text-to-speech · Library coremltools · Repo aufklarer/Chatterbox-Flash-CoreML