Modelpublic

coco-research/coco-voice-models

sourceHugging Faceotherupdated 13d agoView on Hugging Face

0likes242downloads

Model Card

Coco Voice speech models (mirror)

This repository mirrors the speech-to-text model files used by the Coco Voice app that, until now, were hosted only by the Handy project (blob.handy.computer and the handy-computer Hugging Face org).

—Every file is a byte-for-byte copy of Handy's file. Nothing was re-converted, re-quantized or repacked. The SHA-256 of each file is listed below and in its ATTRIBUTION.md, and it equals the Git LFS object id on this repo.
—Layout: <model-id>/<file>, with <model-id>/LICENSE (upstream model license plus Handy's MIT notice) and <model-id>/ATTRIBUTION.md (original publisher, Handy's role, upstream URL, SHA-256) beside each file.
—Licenses differ per model. The per-folder LICENSE and ATTRIBUTION.md govern each model; the repo-level license: other tag just means "see each folder".
—Credits: every model belongs to its original publisher (column below). Handy (MIT, Copyright (c) 2025 CJ Pais) converted, quantized or packaged these files. NVIDIA models are CC-BY-4.0: they are credited to NVIDIA Corporation, and the conversion is indicated in their folders.
—Pin a commit when downloading (/resolve/<commit>/<model-id>/<file>) and verify the SHA-256.

Mirrored models (52 files, 59,191,683,971 bytes, about 59.2 GB)

model id	file	size (bytes)	sha256	license	original publisher	converted by Handy?
`medium`	`whisper-medium-q4_1.bin`	491,852,915 (491.9 MB)	`79283fc1f9fe12ca3248543fbd54b73292164d8df5a16e095e2bceeaaabddf57`	MIT	OpenAI	Yes: ggml q4_1 quant
`breeze-asr`	`breeze-asr-q5_k.bin`	1,080,732,108 (1.08 GB)	`8efbf0ce8a3f50fe332b7617da787fb81354b358c288b008d3bdef8359df64c6`	Apache-2.0	MediaTek Research	Yes: ggml conversion + q5_K quant
`moonshine-tiny-streaming-en`	`moonshine-tiny-streaming-en.tar.gz`	33,006,044 (33.0 MB)	`465addcfca9e86117415677dfdc98b21edc53537210333a3ecdb58509a80abaf`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Repackaged (tar.gz of ORT export)
`moonshine-small-streaming-en`	`moonshine-small-streaming-en.tar.gz`	104,842,676 (104.8 MB)	`dbb3e1c1832bd88a4ac712f7449a136cc2c9a18c5fe33a12ed1b7cb1cfe9cdd5`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Repackaged (tar.gz of ORT export)
`moonshine-medium-streaming-en`	`moonshine-medium-streaming-en.tar.gz`	201,780,020 (201.8 MB)	`07a66f3bff1c77e75a2f637e5a263928a08baae3c29c4c053fc968a9a9373d13`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Repackaged (tar.gz of ORT export)
`canary-180m-flash`	`canary-180m-flash-Q8_0.gguf`	218,447,552 (218.4 MB)	`e13c7f5d0952b056a027cfffec13e3a3a134d1608babed24f983568f141e297c`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q8_0 quant
`cohere-transcribe-03-2026`	`cohere-transcribe-03-2026-Q5_K_M.gguf`	1,770,270,208 (1.77 GB)	`14d02f1ad6dd77b3a60f82639879012c3adb4fe25c50a5a47a2c4c661daf1558`	Apache-2.0	Cohere Labs (Cohere)	Yes: GGUF conversion + Q5KM quant
`whisper-medium`	`whisper-medium-Q8_0.gguf`	831,538,144 (831.5 MB)	`09e6a65e7de377aa5b10bae24608bc6f8ca2ed04b3993ef10d4a02bcd9a82adf`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`Voxtral-Mini-4B-Realtime-2602`	`Voxtral-Mini-4B-Realtime-2602-Q5_K_M.gguf`	3,281,439,008 (3.28 GB)	`e20a7582c5cf8159454c909b7f57b184b287cb3d7c3ec85744727a2cec51b08f`	Apache-2.0	Mistral AI	Yes: GGUF conversion + Q5KM quant
`parakeet-tdt-0.6b-v3`	`parakeet-tdt-0.6b-v3-Q8_0.gguf`	739,508,576 (739.5 MB)	`5859f77944efcd8eafa23a6350731960b2b55b2203df51f319665c807d802cc7`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q8_0 quant
`parakeet-tdt-0.6b-v2`	`parakeet-tdt-0.6b-v2-Q8_0.gguf`	729,574,912 (729.6 MB)	`f0d0e99cebb6d3b83f1f7069b82b5d3c2e39a54545b0da039cb4bafd9c4e5caa`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q8_0 quant
`Qwen3-ASR-0.6B`	`Qwen3-ASR-0.6B-Q8_0.gguf`	850,423,456 (850.4 MB)	`f081b2d5e23bd669d92cc331d722a8a0681943b8e6f34b48996fd5c319b5acd8`	Apache-2.0	Qwen team, Alibaba Cloud	Yes: GGUF conversion + Q8_0 quant
`Fun-ASR-MLT-Nano-2512`	`Fun-ASR-MLT-Nano-2512-Q8_0.gguf`	891,271,232 (891.3 MB)	`d12476d8d9f2baa0ebf738fa955fa05ed33a654f1567289033a810c45d9d9002`	Apache-2.0	FunAudioLLM (Tongyi Lab, Alibaba Group)	Yes: GGUF conversion + Q8_0 quant
`canary-1b-flash`	`canary-1b-flash-Q5_K_M.gguf`	769,563,424 (769.6 MB)	`7eed3cac92f255a4adbd518c58663d3fbf65984d2619189e593f2d374b05c601`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`canary-1b-v2`	`canary-1b-v2-Q5_K_M.gguf`	836,664,032 (836.7 MB)	`9c3a893c93795438baf9b4b1c853c39b60316c3a0d259a3ba6e284712f5ddb71`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`canary-qwen-2.5b`	`canary-qwen-2.5b-Q5_K_M.gguf`	1,983,729,024 (1.98 GB)	`d71b1d9ae442dc538665ed4df63a69520eb9be769ec0bc78c7d2f40040928306`	CC-BY-4.0 (+ Apache-2.0 for the Qwen3-1.7B component)	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`Fun-ASR-Nano-2512`	`Fun-ASR-Nano-2512-Q8_0.gguf`	891,270,912 (891.3 MB)	`681caef6df15a2c0e153b40ca7fe4087fdf65751fa5e6fe605d8a75dff969e61`	Apache-2.0	FunAudioLLM (Tongyi Lab, Alibaba Group)	Yes: GGUF conversion + Q8_0 quant
`gigaam-v3-ctc`	`gigaam-v3-ctc-Q8_0.gguf`	271,803,328 (271.8 MB)	`71e5c82890e9e243a6bd7575f129f5d1bd2c3ca3ae79aab75cfb1a6934c6a62b`	MIT	ai-sage / GigaChat Team (Sber)	Yes: GGUF conversion + Q8_0 quant
`gigaam-v3-e2e-ctc`	`gigaam-v3-e2e-ctc-Q8_0.gguf`	272,151,136 (272.2 MB)	`9ccce4750dc813a493d96ca15ee251712bedec15ac9a02fa3d2bd732f08ae5eb`	MIT	ai-sage / GigaChat Team (Sber)	Yes: GGUF conversion + Q8_0 quant
`gigaam-v3-rnnt`	`gigaam-v3-rnnt-Q8_0.gguf`	273,022,880 (273.0 MB)	`130d54450be2922144b265d90d453b781fc2b5adaee52a16954b00f3686201f7`	MIT	ai-sage / GigaChat Team (Sber)	Yes: GGUF conversion + Q8_0 quant
`gigaam-v3-e2e-rnnt`	`gigaam-v3-e2e-rnnt-Q8_0.gguf`	273,724,832 (273.7 MB)	`78d63b47723b7f8d78c6113a6ef983b5a86e2a86f6c273e1f5cb6967b1c4467a`	MIT	ai-sage / GigaChat Team (Sber)	Yes: GGUF conversion + Q8_0 quant
`granite-speech-4.1-2b-nar`	`granite-speech-4.1-2b-nar-Q5_K_M.gguf`	1,782,089,344 (1.78 GB)	`88d7c7b5b8b59c95bb6580a1e7d5d81cae63943cef71405ff477527b0bb69fca`	Apache-2.0	IBM (IBM Granite team)	Yes: GGUF conversion + Q5KM quant
`granite-4.0-1b-speech`	`granite-4.0-1b-speech-Q5_K_M.gguf`	1,829,704,640 (1.83 GB)	`4f47208116b3d4dcf73b8562a739bd675f80eeefb4f64e7481b08aee223c9b86`	Apache-2.0	IBM (IBM Granite team)	Yes: GGUF conversion + Q5KM quant
`granite-speech-4.1-2b`	`granite-speech-4.1-2b-Q5_K_M.gguf`	1,829,704,640 (1.83 GB)	`1ea15c2174b485b8de81337a1da3ae6703a7892a5330345b9a2212e9a6a0c365`	Apache-2.0	IBM (IBM Granite team)	Yes: GGUF conversion + Q5KM quant
`granite-speech-4.1-2b-plus`	`granite-speech-4.1-2b-plus-Q5_K_M.gguf`	1,691,297,216 (1.69 GB)	`4b762cb221169db53863aa2ace3489310d619263b3a5382b39f1abac60f87301`	Apache-2.0	IBM (IBM Granite team)	Yes: GGUF conversion + Q5KM quant
`moonshine-tiny`	`moonshine-tiny-Q8_0.gguf`	35,466,912 (35.5 MB)	`2fd348d7b38f97d309cc3ec6848f3f57f537b80244950f07d2637e463f95a3a1`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Yes: GGUF conversion + Q8_0 quant
`moonshine-streaming-tiny`	`moonshine-streaming-tiny-Q8_0.gguf`	50,462,816 (50.5 MB)	`930e4622ad3a24158b91406c30c977fa6a26b34cb32d6ac3e57cfb23383a869e`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Yes: GGUF conversion + Q8_0 quant
`moonshine-base`	`moonshine-base-Q8_0.gguf`	77,476,480 (77.5 MB)	`7f0027dfd857d310b63a85ef57cadf183da712cc374f85a648f8bc18aaa2efc8`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Yes: GGUF conversion + Q8_0 quant
`moonshine-streaming-small`	`moonshine-streaming-small-Q8_0.gguf`	198,506,848 (198.5 MB)	`d03670f69629b649085d0f44a63d97668b4119117cc9611a4e4ad94341713dfc`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Yes: GGUF conversion + Q8_0 quant
`moonshine-streaming-medium`	`moonshine-streaming-medium-Q8_0.gguf`	295,793,568 (295.8 MB)	`f7c9564249b508f6012927ec4f9e536087da53a7047f858ca9975bea5f75299e`	MIT	Useful Sensors, Inc. (dba Moonshine AI)	Yes: GGUF conversion + Q8_0 quant
`parakeet-tdt_ctc-110m`	`parakeet-tdt_ctc-110m-Q8_0.gguf`	135,373,280 (135.4 MB)	`7dd44c74a331d788a4e5f8b16913b3feb29ced22cf5613aad0e0f6cd30516296`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q8_0 quant
`parakeet-ctc-0.6b`	`parakeet-ctc-0.6b-Q8_0.gguf`	722,271,424 (722.3 MB)	`e47ae86c1b34dbaf054334592575c29aee7de3546e75d1f0ccee05904c5c49ae`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q8_0 quant
`parakeet-rnnt-0.6b`	`parakeet-rnnt-0.6b-Q8_0.gguf`	729,687,456 (729.7 MB)	`06da5ba773736984e0513037b27ff9300e95c55eaa2b79dbc688efc34b25df51`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q8_0 quant
`parakeet-ctc-1.1b`	`parakeet-ctc-1.1b-Q5_K_M.gguf`	928,584,736 (928.6 MB)	`8b38f60bb1c56b518953a39703068adc09991993b5aab52782e038a86a6c04cc`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`parakeet-tdt-1.1b`	`parakeet-tdt-1.1b-Q5_K_M.gguf`	935,758,496 (935.8 MB)	`18c22888fed10676fde72fdd5b833fab9175a6abf4e0cb390eb0a59b0b107cd3`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`parakeet-rnnt-1.1b`	`parakeet-rnnt-1.1b-Q5_K_M.gguf`	935,755,008 (935.8 MB)	`8ee2ead0c30773440743591cd8a9f8ac180c27f450c9b053805b2bab6bf352cb`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`parakeet-tdt_ctc-1.1b`	`parakeet-tdt_ctc-1.1b-Q5_K_M.gguf`	935,758,080 (935.8 MB)	`da23eb8c78fe3813b88f616db85e8191cfbfffdbc2b8bb8bd74a94eb8987caee`	CC-BY-4.0	NVIDIA Corporation	Yes: GGUF conversion + Q5KM quant
`Qwen3-ASR-1.7B`	`Qwen3-ASR-1.7B-Q5_K_M.gguf`	1,517,290,464 (1.52 GB)	`034c557fe92ff8fcd9a9c041cbdaad347be0a86a58d3a348f63cf3f0180879d0`	Apache-2.0	Qwen team, Alibaba Cloud	Yes: GGUF conversion + Q5KM quant
`Voxtral-Mini-3B-2507`	`Voxtral-Mini-3B-2507-Q5_K_M.gguf`	3,464,182,432 (3.46 GB)	`9ea2eb85359f7771270715849dc2b0e80381a59dab09d9c5d85352ab9c53f675`	Apache-2.0	Mistral AI	Yes: GGUF conversion + Q5KM quant
`Voxtral-Small-24B-2507`	`Voxtral-Small-24B-2507-Q5_K_M.gguf`	17,138,659,808 (17.14 GB)	`a53f73a5f63b7663fe155977616a636acac567dac898f6ef82ca519017e8b6e7`	Apache-2.0	Mistral AI	Yes: GGUF conversion + Q5KM quant
`whisper-tiny`	`whisper-tiny-Q8_0.gguf`	45,981,088 (46.0 MB)	`325b9c7997cd1eff81ef709d55766565e71be696130cc3a3d444713798706834`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-tiny.en`	`whisper-tiny.en-Q8_0.gguf`	45,904,544 (45.9 MB)	`e8c9b73c06344307d8b346e07fbe93dd88d894627854bcff31523f1ce44394fa`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-base`	`whisper-base-Q8_0.gguf`	84,962,880 (85.0 MB)	`81c069428bc8a24551a8169cf31cf09bcfd9d4cf50389ae281323c9aa9648c81`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-base.en`	`whisper-base.en-Q8_0.gguf`	84,886,208 (84.9 MB)	`3b46ca40bccbf7609c68d88a36d96077a04ca7c87f2060ede06f129fac3e7652`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-small.en`	`whisper-small.en-Q8_0.gguf`	269,674,144 (269.7 MB)	`9614e6b7fda2d26018e4f268aece8ca25a83296ea0b534169a585b740bfd71ef`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-small`	`whisper-small-Q8_0.gguf`	269,751,136 (269.8 MB)	`9b9c8811bbcc82a7766f0fb0925614bdacb0923b2cc630daeac17108b655b860`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-medium.en`	`whisper-medium.en-Q8_0.gguf`	831,460,928 (831.5 MB)	`03d7257fef498750ce272631bc6a34de322fc2b438aab5c268ff49dfd1b64c49`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-large-v3-turbo`	`whisper-large-v3-turbo-Q8_0.gguf`	886,381,760 (886.4 MB)	`b2e30cc286bc9f3aba4db9099fc7403543497c05ce7100d0d83091ddfd25a183`	MIT	OpenAI	Yes: GGUF conversion + Q8_0 quant
`whisper-large-v3`	`whisper-large-v3-Q5_K_M.gguf`	1,161,143,008 (1.16 GB)	`6053d0fd69a0fd48b8fea5ea7a52b9e0cde389343566fa30e453a1b2b258dc38`	MIT	OpenAI	Yes: GGUF conversion + Q5KM quant
`whisper-large`	`whisper-large-Q5_K_M.gguf`	1,160,366,048 (1.16 GB)	`b1dfb9fa0e9b0574d8c822f51662b8e59915831e1a3ca0d191f18ff8ba742639`	MIT	OpenAI	Yes: GGUF conversion + Q5KM quant
`whisper-large-v2`	`whisper-large-v2-Q5_K_M.gguf`	1,160,366,080 (1.16 GB)	`94fab1289a98d26bf375174a8daa00f44c7b16a9944726dde05361e5a8bd07ff`	MIT	OpenAI	Yes: GGUF conversion + Q5KM quant
`Breeze-ASR-25`	`Breeze-ASR-25-Q5_K_M.gguf`	1,160,366,080 (1.16 GB)	`c871fc811b33a16a5607c4d4166cfa2c0a1d359f7796e3da255e3e922f59139b`	Apache-2.0	MediaTek Research	Yes: GGUF conversion + Q5KM quant

Model ids are the Coco Voice ids: catalog.json slugs for the GGUF entries, and model.rs ids for medium, breeze-asr and the three moonshine-*-streaming-en tarballs. Some catalog slugs, such as parakeet-tdt-0.6b-v3, are also used by older model.rs entries. Those older entries are not hosted here (see the next section); here the folder holds the catalog GGUF.

Not hosted here: publisher-direct models

The following Coco Voice models are not hosted in this repository. The app downloads them directly from the original publisher's (or the original converter's) Hugging Face repository, pinned to a commit, because those files are byte-identical (same SHA-256) to what Handy shipped:

Coco Voice model id (model.rs)	downloaded from (repo @ pinned commit)
`small`	ggerganov/whisper.cpp@`5359861`
`turbo`	ggerganov/whisper.cpp@`5359861`
`large`	ggerganov/whisper.cpp@`5359861`
`parakeet-tdt-0.6b-v2`	istupakov/parakeet-tdt-0.6b-v2-onnx@`0bbb45a`
`parakeet-tdt-0.6b-v3`	istupakov/parakeet-tdt-0.6b-v3-onnx@`8f23f0c`
`moonshine-base`	moonshine-ai/moonshine@`48b4e42`
`sense-voice-int8`	csukuangfj/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17@`2365bae`
`gigaam-v3-e2e-ctc`	istupakov/gigaam-v3-onnx@`322c3b2`
`canary-180m-flash`	istupakov/canary-180m-flash-onnx@`92c2231`, istupakov/parakeet-tdt-0.6b-v3-onnx@`8f23f0c`
`canary-1b-v2`	istupakov/canary-1b-v2-onnx@`5ebc152`, istupakov/parakeet-tdt-0.6b-v3-onnx@`8f23f0c`
`cohere-int8`	csukuangfj2/sherpa-onnx-cohere-transcribe-14-lang-int8-2026-04-01@`156a470`, tristanripke/cohere-transcribe-onnx-int8@`9ecc3a5`

Some other models that earlier Handy builds offered are also intentionally not mirrored here, for licensing reasons.

Disclaimer

The models are provided as-is, under their respective licenses, without warranty of any kind. Coco Research is not affiliated with, or endorsed by, the original publishers or the Handy project.