schema_version: 1
release: release-0.5.1

# Verified from Hugging Face file metadata on 2026-08-13. These are offline UI
# estimates; live Content-Length values remain authoritative during downloads.
package_sizes:
  citrinet_asr_q8_0: 40574432
  chatterbox_f16: 3744360386
  chatterbox_q8_0: 2088393668
  chatterbox_safetensors: 3191859618
  confucius4_tts_orig: 8192757760
  dramabox_q8_0: 18942803808
  fish_audio_s2_pro_bf16: 10229278080
  fish_audio_s2_pro_q8_0: 6317911232
  fun_asr_nano_2512_f16: 1675708832
  fun_asr_nano_2512_q8_0: 1045334432
  fun_asr_nano_2512_safetensors: 1671205126
  glm_tts_q8_0: 5143813728
  higgs_audio_tts_4b_bf16: 8501587648
  higgs_audio_tts_4b_q8_0: 5095354048
  higgs_audio_stt_f16: 5367453248
  higgs_audio_stt_q8_0: 3158310848
  hviske_asr_q8_0: 2436359808
  hviske_asr_safetensors: 4132293306
  index_tts2_f16: 4646898304
  index_tts2_orig: 8084552000
  index_tts2_q8_0: 3633888608
  index_tts2_safetensors: 3484956803
  inflect_micro_v2_orig: 72082176
  irodori_tts_500m_v3_f16: 1254813120
  irodori_tts_500m_v3_q8_0: 1093739584
  irodori_tts_600m_v3_voicedesign_f16: 1463787680
  irodori_tts_600m_v3_voicedesign_q8_0: 1272140064
  irodori_tts_v4_small_f16: 1762148352
  irodori_tts_v4_small_q8_0: 1368991360
  kroko_asr_community_q8_0: 167756928
  miotts_1_7b_bf16: 3518532512
  miotts_1_7b_orig: 3518532512
  miotts_1_7b_q8_0: 2197326752
  moss_tts_local_v1_5_bf16: 13367797280
  moss_tts_local_v1_5_q8_0: 7512220768
  moss_tts_nano_100m_bf16: 332423040
  moss_tts_nano_100m_q8_0: 193337984
  nemotron_asr_f16: 1277710880
  nemotron_asr_q8_0: 930620256
  nemotron_asr_safetensors: 2552818890
  omnivoice_bf16: 1639548640
  omnivoice_f16: 1639548768
  omnivoice_q8_0: 1350288416
  omnivoice_safetensors: 2461770336
  outetts_1_0_1b_q8_0: 3029895456
  parakeet_tdt_f16: 1255384320
  parakeet_tdt_q8_0: 915733744
  pocket_tts_english_bf16: 219096064
  pocket_tts_english_q8_0: 127856704
  pocket_tts_english_safetensors: 385107895
  pocket_tts_german_bf16: 219096544
  pocket_tts_german_q8_0: 127857184
  pocket_tts_italian_bf16: 219096800
  pocket_tts_italian_q8_0: 127857440
  pocket_tts_portuguese_bf16: 219097728
  pocket_tts_portuguese_q8_0: 127858368
  pocket_tts_spanish_bf16: 219097600
  pocket_tts_spanish_q8_0: 127858240
  qwen3_tts_0_6b_base_safetensors: 2511651783
  qwen3_tts_1_7b_base_bf16: 4203158464
  qwen3_tts_1_7b_base_orig: 4544273280
  qwen3_tts_1_7b_base_q8_0: 2695175104
  qwen3_tts_1_7b_base_safetensors: 4539721255
  qwen3_tts_1_7b_customvoice_bf16: 4179144352
  qwen3_tts_1_7b_customvoice_q8_0: 2817044064
  qwen3_tts_1_7b_voicedesign_bf16: 4179089248
  qwen3_tts_1_7b_voicedesign_q8_0: 2816988960
  qwen3_asr_0_6b_f16: 1880642016
  qwen3_asr_0_6b_q8_0: 1151272416
  qwen3_asr_0_6b_safetensors: 1876110856
  qwen3_asr_1_7b_f16: 4087653248
  qwen3_asr_1_7b_q8_0: 2473010048
  qwen3_asr_1_7b_safetensors: 4087626782
  seed_vc_mlx_f16: 3629186560
  seed_vc_mlx_orig: 7003311936
  seed_vc_mlx_q8_0: 3120809248
  seed_vc_mlx_safetensors: 712430273
  supertonic_3_f16: 312784196
  supertonic_3_orig: 454072836
  supertonic_3_q8_0: 454072836
  supertonic_3_safetensors: 397279577
  vevo2_f16: 5021112128
  vevo2_orig: 7449615488
  vevo2_q8_0: 3242124800
  vibevoice_1_5b_bf16: 5420021858
  vibevoice_1_5b_q8_0: 3224701538
  vibevoice_asr_f16: 17361090304
  vibevoice_asr_q8_0: 9858644224
  vietneu_tts_v3_turbo_q8_0: 170482368
  voxcpm2_bf16: 4772288288
  voxcpm2_orig: 4960716192
  voxcpm2_q8_0: 2955000480
  voxcpm2_safetensors: 4583766759
  voxtral_realtime_bf16: 8874402784
  voxtral_realtime_q4_k: 3097662432
  voxtral_realtime_q8_0: 5104567264

# Runtime dependencies omitted by upstream TTS package declarations. These are
# installed independently and passed to audio.cpp through absolute session paths.
package_dependencies:
  miotts_1_7b_q8_0: &miotts_codec_dependency
    - id: miocodec_q8_0
      family: miocodec
      display_name: "MioCodec 25Hz 44.1kHz v2 Q8_0 GGUF"
      target_directory: "MioCodec-25Hz-44.1kHz-v2-GGUF"
      format: gguf
      precision: q8_0
      files: ["MioCodec-25Hz-44.1kHz-v2-GGUF/miocodec-25hz-44khz-v2-q8_0.gguf"]
      strip_prefix: "MioCodec-25Hz-44.1kHz-v2-GGUF"
      repo: "audio-cpp/audio.cpp-gguf"
      revision: main
      estimated_download_bytes: 299066464
      session_option: "miotts.codec_model_path"
  miotts_1_7b_bf16: *miotts_codec_dependency
  miotts_1_7b_orig: *miotts_codec_dependency

# Suite-owned UI and validation metadata. Download/package truth remains in
# model_specs/*.json, which are unmodified snapshots of audio.cpp.
defaults:
  suite_support: supported
  suite_tasks: [tts, srt, character_switching]
  reference_audio: none
  reference_transcript: none
  built_in_voices: false
  voice_design: false
  inline_controls: false
  native_multi_speaker: {supported: false, max_speakers: 1, suite_status: unavailable}
  asr_features: {diarization: none, timing: none}

families:
  citrinet_asr:
    suite_tasks: [asr]
    summary: "Compact English transcription using a Citrinet CTC model."
  chatterbox:
    reference_audio: required
    suite_tasks: [tts, srt, character_switching, voice_conversion]
    summary: "Voice cloning and reference-targeted voice conversion."
  confucius4_tts: {reference_audio: required, summary: "Voice cloning from reference audio."}
  dramabox: {reference_audio: optional, summary: "TTS and optional voice cloning."}
  fish_audio: {reference_audio: optional, summary: "Reference-conditioned TTS."}
  fun_asr_nano:
    suite_tasks: [asr]
    summary: "Multilingual speech recognition for Chinese, English, and Japanese."
  glm_tts:
    reference_audio: required
    reference_transcript: required
    summary: "Zero-shot cloning; reference audio and its transcript are required."
  higgs_audio_tts:
    reference_audio: optional
    reference_transcript: optional
    inline_controls: true
    summary: "Expressive multilingual TTS, cloning, and inline style/sound controls."
  higgs_audio_stt:
    suite_tasks: [asr]
    summary: "English transcription using the Higgs Audio v3 speech model."
  hviske_asr:
    suite_tasks: [asr]
    summary: "Multilingual transcription across 16 languages."
  index_tts2: {reference_audio: optional, summary: "TTS with optional voice cloning."}
  inflect_v2: {summary: "Reference-free expressive TTS."}
  irodori_tts:
    reference_audio: optional
    voice_design: true
    summary: "TTS, optional cloning, and voice design."
  kroko_asr:
    suite_tasks: [asr]
    asr_features: {timing: native_word}
    summary: "Small multilingual community ASR model."
  miotts: {reference_audio: optional, summary: "Reference-conditioned TTS."}
  moss_tts_local: {reference_audio: optional, summary: "TTS with optional voice cloning."}
  moss_tts_nano: {reference_audio: optional, summary: "Compact TTS with optional cloning."}
  nemotron_asr:
    suite_tasks: [asr]
    asr_features: {timing: native_word}
    summary: "Streaming-capable multilingual NVIDIA Nemotron transcription."
  omnivoice:
    reference_audio: optional
    reference_transcript: optional
    voice_design: true
    summary: "Reference cloning or instruction-based voice design."
  outetts:
    reference_audio: optional
    reference_transcript: optional
    summary: "Multilingual TTS; a transcript improves reference cloning."
  parakeet_tdt:
    suite_tasks: [asr]
    asr_features: {timing: native_word}
    summary: "Fast multilingual Parakeet-TDT transcription."
  pocket_tts: {reference_audio: optional, summary: "Small CPU-friendly TTS and cloning."}
  qwen3_tts:
    reference_audio: optional
    voice_design: true
    summary: "TTS with package-dependent cloning or voice design."
  qwen3_asr:
    suite_tasks: [asr]
    asr_features: {timing: optional_forced_aligner}
    summary: "Broad multilingual Qwen3 transcription; word timestamps require an optional forced aligner."
  seed_vc:
    suite_tasks: [voice_conversion]
    reference_audio: required
    summary: "Reference-targeted voice conversion through Seed-VC."
  supertonic:
    built_in_voices: true
    summary: "Fast multilingual TTS using built-in voices."
  vevo2:
    reference_audio: optional
    suite_tasks: [tts, srt, character_switching, voice_conversion]
    summary: "TTS and reference-targeted voice conversion; S2S/SVC remain advanced upstream routes."
  vibevoice:
    reference_audio: required_per_speaker
    native_multi_speaker: {supported: true, max_speakers: 4, suite_status: partial}
    summary: "Long-form dialogue for up to four speakers. Native single-request mode is not wired yet."
  vibevoice_asr:
    suite_tasks: [asr, diarization]
    asr_features: {diarization: native, timing: native_segment}
    summary: "Multilingual transcription with native timestamped speaker turns."
  vietneu_tts:
    voice_design: true
    summary: "Vietnamese TTS and voice design."
  voxcpm2:
    reference_audio: optional
    voice_design: true
    summary: "TTS with reference conditioning and voice design."
  voxtral_realtime:
    suite_tasks: [asr]
    summary: "Multilingual realtime-oriented Voxtral transcription."