audio.cpp-gguf-fork

This directory contains audio.cpp-native GGUF conversions of multiple speech models. These files are intended for use with [audio.cpp](https://github.com/0xShug0/audio.cpp). For conversion details, supported layouts, direct-file loading, sidecar embedding, and the latest compatibility notes, see the audio.cpp GGUF guide: - https://github.com/0xShug0/audio.cpp/blob/main/docs/gguf.md !!! Automated audio checks are intentionally strict and may flag length, log-mel, or transcript drift that can still sound acceptable to human listeners. Validate the exact file, backend, and route you plan to use. These converted weights are provided as-is; use them at your own risk. `Tested` summarizes the current audio.cpp path-test status. See the GGUF guide above for the full matrix and drift notes. - Chatterbox Q8 is intentionally mixed type. Graph-sensitive scalar, norm, bias, and side tensors stay in non-Q8 types while matmul-compatible weights are quantized. - PocketTTS Q8 keeps the four `flow_lm.flow_net.time_embed.*.mlp.{0,2}.weight` tensors in Q8 in addition to the default converter selection. `conditioner.embed`, `cond_embed`, and Mimi conv tensors are not forced to Q8 because tested outputs drifted or the current conv path casts quantized conv weights back to F32. - Voxtral Q4_K is smaller than Q8_0 and was faster in a quick CUDA path check, with transcripts matching Q8_0 except for one capitalization-only difference. Pass a GGUF file directly as `--model`: audiocpp_cli --task tts --family supertonic --model Supertonic-3-GGUF/supertonic-3-orig.gguf --backend cuda --language en --text "Hello." --voice-id M1 --out out.wav For ASR: audiocpp_cli --task asr --family qwen3_asr --model Qwen3-ASR-0.6B-GGUF/qwen3-asr-0.6b-f16.gguf --backend cuda --audio speech.wav --text "" --text-out transcript.txt Each GGUF file is a converted form of its original model. Use and redistribution are governed by the corresponding original model license listed above. Please review the original model card and license terms before using or redistributing any converted weights.