{ "cases": [ { "id": "ace_step_turbo_text2music_long_chunk", "coverage": "ACE-Step turbo text2music, planner, DiT, long VAE decode chunking, then shorter request in the same session", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "long_160s", "text": "expansive synth pop with bright vocals and steady drums", "language": "en", "task_route": "text2music", "lyrics": "[verse]\nNeon rivers cross the sky\n[chorus]\nWe keep dancing through the night", "duration_seconds": 160, "options": { "lm_temperature": 0.0, "lm_top_k": 0, "lm_top_p": 1.0 }, "seed": 1234, "guidance_scale": 1.0, "num_inference_steps": 8 }, { "id": "short_20s", "text": "gentle piano pop with intimate vocals", "language": "en", "task_route": "text2music", "lyrics": "[verse]\nSoft rain taps the window\n[chorus]\nHold the morning close", "duration_seconds": 20, "options": { "lm_temperature": 0.0, "lm_top_k": 0, "lm_top_p": 1.0 }, "seed": 1234, "guidance_scale": 1.0, "num_inference_steps": 8 } ] }, { "id": "ace_step_base_cfg_text2music", "coverage": "ACE-Step base CFG text2music route", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "load_options": { "ace_step.dit_model_path": "acestep-v15-base", "ace_step.lm_model_path": "acestep-5Hz-lm-1.7B" }, "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "base_cfg_text2music", "text": "energetic chiptune J-pop with a bright female vocal", "language": "ja", "task_route": "text2music", "lyrics": "[verse]\n森の奥で光る声\n[chorus]\n月夜に響く君の歌", "duration_seconds": 20, "options": { "lm_cfg_scale": 2.0, "lm_temperature": 0.0, "lm_top_k": 0, "lm_top_p": 1.0 }, "seed": 1234, "guidance_scale": 7.0, "num_inference_steps": 50 } ] }, { "id": "ace_step_complete_source_audio", "coverage": "ACE-Step base complete route with source audio", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "load_options": { "ace_step.dit_model_path": "acestep-v15-base", "ace_step.lm_model_path": "acestep-5Hz-lm-1.7B" }, "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "complete_source_audio", "text": "complete this song with clear vocals and a fuller pop arrangement", "language": "en", "audio": "resources/mel_roformer_test_10s.wav", "task_route": "complete", "duration_seconds": 10, "options": { "lm_cfg_scale": 2.0, "lm_temperature": 0.0 }, "seed": 5701, "guidance_scale": 7.0, "num_inference_steps": 30 } ] }, { "id": "ace_step_lego_guitar_from_piano", "coverage": "ACE-Step base lego route with piano source audio", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "load_options": { "ace_step.dit_model_path": "acestep-v15-base", "ace_step.lm_model_path": "acestep-5Hz-lm-1.7B" }, "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "lego_guitar_from_piano", "text": "add a clean guitar track that follows the source piano", "language": "en", "audio": "resources/ace_step_lego_piano_10s.wav", "task_route": "lego", "track_name": "guitar", "duration_seconds": 10, "options": { "lm_cfg_scale": 2.0, "lm_temperature": 0.0 }, "seed": 5702, "guidance_scale": 7.0, "num_inference_steps": 50 } ] }, { "id": "ace_step_extract_vocals", "coverage": "ACE-Step base extract vocal route", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "load_options": { "ace_step.dit_model_path": "acestep-v15-base", "ace_step.lm_model_path": "acestep-5Hz-lm-1.7B" }, "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "extract_vocals", "text": "vocals", "language": "en", "audio": "resources/mel_roformer_test_15s.wav", "task_route": "extract", "track_name": "vocals", "duration_seconds": 15, "options": { "lm_cfg_scale": 2.0, "lm_temperature": 0.0, "lm_top_k": 0, "lm_top_p": 1.0 }, "seed": 1234, "num_inference_steps": 50 } ] }, { "id": "ace_step_cover_nofsq", "coverage": "ACE-Step base cover-nofsq route", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "load_options": { "ace_step.dit_model_path": "acestep-v15-base", "ace_step.lm_model_path": "acestep-5Hz-lm-1.7B" }, "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "cover_nofsq", "text": "make a bright pop cover with a new vocal tone", "language": "en", "audio": "resources/mel_roformer_test_10s.wav", "task_route": "cover-nofsq", "duration_seconds": 10, "options": { "lm_cfg_scale": 2.0, "lm_temperature": 0.0 }, "seed": 5704, "guidance_scale": 7.0, "num_inference_steps": 50 } ] }, { "id": "ace_step_repaint_middle", "coverage": "ACE-Step base repaint route with source audio splice", "family": "ace_step", "model": "models/Ace-Step1.5", "task": "tts", "mode": "offline", "load_options": { "ace_step.dit_model_path": "acestep-v15-base", "ace_step.lm_model_path": "acestep-5Hz-lm-1.7B" }, "session_options": { "ace_step.weight_type": "f32" }, "outputs": [ "audio" ], "requests": [ { "id": "repaint_middle", "text": "replace the middle section with a stronger vocal phrase", "language": "en", "audio": "resources/mel_roformer_test_15s.wav", "task_route": "repaint", "duration_seconds": 15, "repaint_start": 3.0, "repaint_end": 8.0, "repaint_mode": "balanced", "repaint_strength": 0.6, "options": { "lm_cfg_scale": 2.0, "lm_temperature": 0.0 }, "seed": 5705, "guidance_scale": 7.0, "num_inference_steps": 50 } ] }, { "id": "qwen3_tts_base_voice_clone_reuse", "coverage": "Qwen3 TTS base text generation, voice clone prompt encoding, talker/decoder reuse across long-short-long requests", "family": "qwen3_tts", "model": "models/Qwen3-TTS-12Hz-1.7B-Base", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "long", "text": "The lighthouse keeper records a detailed morning report about the tide, the lantern, and the ships crossing the horizon.", "language": "English", "voice_ref": "resources/a.wav", "reference_text": "This little work was finished in the year eighteen o three, and intended for immediate publication.", "seed": 1234, "max_tokens": 256, "do_sample": false }, { "id": "short", "text": "The shorter line should keep the same speaker.", "language": "English", "voice_ref": "resources/a.wav", "reference_text": "This little work was finished in the year eighteen o three, and intended for immediate publication.", "seed": 1234, "max_tokens": 128, "do_sample": false }, { "id": "long_again", "text": "After the short request, the same voice describes the harbor again in a longer passage so graph reuse and capacity behavior stay visible.", "language": "English", "voice_ref": "resources/a.wav", "reference_text": "This little work was finished in the year eighteen o three, and intended for immediate publication.", "seed": 1234, "max_tokens": 256, "do_sample": false } ] }, { "id": "qwen3_tts_voice_design", "coverage": "Qwen3 TTS voice-design variant and instruction path", "family": "qwen3_tts", "model": "models/Qwen3-TTS-12Hz-1.7B-VoiceDesign", "task": "vdes", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "design", "text": "This designed voice should sound cheerful and clear while keeping every word easy to recognize.", "language": "English", "instruct": "A cheerful young woman with a bright studio voice.", "seed": 1234, "max_tokens": 192, "do_sample": false } ] }, { "id": "qwen3_tts_custom_voice", "coverage": "Qwen3 TTS custom voice variant speaker selection path", "family": "qwen3_tts", "model": "models/Qwen3-TTS-12Hz-1.7B-CustomVoice", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "custom", "text": "The custom voice request should remain stable for a medium length sentence.", "language": "English", "speaker": "Vivian", "seed": 1234, "max_tokens": 192, "do_sample": false } ] }, { "id": "chatterbox_en_long_cfg", "coverage": "Chatterbox English voice clone, T3, S3 tokenizer, CFM/HiFT, CFG two-batch path, long text", "family": "chatterbox", "model": "models/chatterbox", "task": "clon", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "en_long", "text": "The archivist unlocks the reading room and describes the maps, the weathered journals, and the quiet harbor outside in one steady natural voice.", "language": "en", "voice_ref": "resources/sample.wav", "options": { "s3gen_cfg_rate": 0.7 }, "seed": 1234, "max_tokens": 384, "guidance_scale": 0.5 } ] }, { "id": "chatterbox_de_multilingual", "coverage": "Chatterbox multilingual component selection and non-English frontend path", "family": "chatterbox", "model": "models/chatterbox", "task": "clon", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "de", "text": "Der Leuchtturmwaerter beschreibt den ruhigen Morgen und die Schiffe am Horizont.", "language": "de", "voice_ref": "resources/sample.wav", "options": { "s3gen_cfg_rate": 0.7 }, "seed": 1234, "max_tokens": 256, "guidance_scale": 0.5 } ] }, { "id": "miocodec_voice_conversion", "coverage": "MioCodec voice conversion path, WavLM feature extraction, content/global encoders, wave decoder, and waveform reconstruction", "family": "miocodec", "model": "models/MioCodec-25Hz-44.1kHz-v2", "task": "vc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "design_to_sample", "audio": "resources/design.wav", "voice_ref": "resources/sample.wav" } ] }, { "id": "miotts_long_text_voice_clone", "coverage": "MioTTS voice clone path, long text chunking, causal LM sampling, MioCodec token decode, and waveform reconstruction", "family": "miotts", "model": "models/MioTTS-1.7B", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "long_clone", "text": "For audit, this MioTTS path test uses a passage that is longer than the default chunk size but still easy to verify. The first sentence checks that the opening words are spoken clearly. The second sentence says that green clocks measure quiet rivers while silver lanterns mark the path home. The final sentence confirms that the ending phrase is present and easy to hear.", "voice_ref": "resources/sample.wav", "seed": 1234, "max_tokens": 1400, "do_sample": true } ] }, { "id": "voxcpm2_tts_voice_design", "coverage": "VoxCPM2 text-to-speech path with text-embedded voice design, MiniCPM generation, diffusion feature generation, and AudioVAE decode", "family": "voxcpm2", "model": "models/VoxCPM2", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "voice_design", "text": "(A calm studio narrator with clear articulation) This VoxCPM2 path test checks the text-to-speech interface through AudioCPP CLI.", "seed": 1234, "max_tokens": 0, "guidance_scale": 2.0, "num_inference_steps": 10 } ] }, { "id": "voxcpm2_voice_clone", "coverage": "VoxCPM2 voice clone path with reference audio encoding, MiniCPM generation, diffusion feature generation, and AudioVAE decode", "family": "voxcpm2", "model": "models/VoxCPM2", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "clone", "text": "This VoxCPM2 path test clones the reference speaker for a short review sentence.", "voice_ref": "resources/sample.wav", "seed": 1234, "max_tokens": 0, "guidance_scale": 2.0, "num_inference_steps": 10 } ] }, { "id": "seed_vc_v2_vc_long_bigvgan", "coverage": "SeedVC V2 voice conversion, AR, CFM, CampPlus, BigVGAN chunked vocoder", "family": "seed_vc", "model": "models/SeedVC-MLX", "task": "vc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "v2_long", "audio": "resources/seed_vc/source/see_you_again_vocals_cut_28sec.wav", "voice_ref": "resources/seed_vc/voice_ref/s2p1.wav", "route": "v2_vc", "options": { "length_adjust": 1.0, "intelligibility_cfg_rate": 0.7, "similarity_cfg_rate": 0.7 }, "seed": 1234, "num_inference_steps": 30 } ] }, { "id": "seed_vc_v1_svc_long_rmvpe_bigvgan", "coverage": "SeedVC V1 SVC, Whisper content, RMVPE F0, length adjust, BigVGAN long chunking", "family": "seed_vc", "model": "models/SeedVC-MLX", "task": "svc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "v1_svc_180s", "audio": "resources/mel_roformer_test_180s.wav", "voice_ref": "resources/seed_vc/voice_ref/s2p1.wav", "route": "v1_svc", "options": { "length_adjust": 1.0, "inference_cfg_rate": 0.7, "f0_condition": true, "auto_f0_adjust": true, "semi_tone_shift": 0, "fp16": false }, "seed": 1234, "num_inference_steps": 30 } ] }, { "id": "seed_vc_v1_whisper_bigvgan_vc", "coverage": "SeedVC V1 whisper-small BigVGAN voice-conversion route", "family": "seed_vc", "model": "models/SeedVC-MLX", "task": "vc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "v1_whisper_bigvgan", "audio": "resources/seed_vc/source/source_s1.wav", "voice_ref": "resources/seed_vc/voice_ref/s2p1.wav", "route": "v1_whisper_bigvgan_vc", "options": { "length_adjust": 1.0, "inference_cfg_rate": 0.7, "fp16": false }, "seed": 1234, "num_inference_steps": 30 } ] }, { "id": "seed_vc_v1_xlsr_hift_vc", "coverage": "SeedVC V1 XLSR/HuBERT-style content encoder and HiFT vocoder route", "family": "seed_vc", "model": "models/SeedVC-MLX", "task": "vc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "v1_xlsr_hift", "audio": "resources/seed_vc/source/source_s1.wav", "voice_ref": "resources/seed_vc/voice_ref/s2p1.wav", "route": "v1_xlsr_hift_vc", "options": { "length_adjust": 1.0, "inference_cfg_rate": 0.7, "fp16": false }, "seed": 1234, "num_inference_steps": 30 } ] }, { "id": "vevo2_tts_routes", "coverage": "VeVo2 text-to-speech task routes: zero-shot speech, text-to-singing, and singing voice synthesis", "family": "vevo2", "model": "models/Vevo2", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "zero_shot_tts", "route": "zero_shot_tts", "target_text": "This is a zero shot speech request using the unified VeVo2 command line interface.", "style_ref": "resources/vevo2_official_gallery/tts_ref.wav", "style_ref_text": "This reference describes the target speaking voice.", "target_voice": "resources/vevo2_official_gallery/tts_ref.wav", "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 }, { "id": "text_to_singing", "route": "text_to_singing", "target_text": "A simple melody carries the words through the evening light.", "style_ref": "resources/vevo2_official_gallery/adele_raw.wav", "style_ref_text": "This reference describes the target singing voice.", "target_voice": "resources/vevo2_official_gallery/adele_raw.wav", "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 }, { "id": "svs", "route": "svs", "target_text": "The chorus rises softly and returns with a brighter tone.", "style_ref": "resources/vevo2_official_gallery/svs_melody.wav", "style_ref_text": "This reference describes the target melody and singing style.", "target_voice": "resources/vevo2_official_gallery/adele_raw.wav", "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 } ] }, { "id": "vevo2_vc_routes", "coverage": "VeVo2 voice-conversion task routes: style-preserved and style-converted voice conversion", "family": "vevo2", "model": "models/Vevo2", "task": "vc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "style_preserved_vc", "route": "style_preserved_vc", "source_audio": "resources/vevo2_official_gallery/style_whisper2normal_input.wav", "target_voice": "resources/vevo2_official_gallery/tts_ref.wav", "use_pitch_shift": false, "seed": 1234, "num_inference_steps": 32 }, { "id": "style_converted_vc", "route": "style_converted_vc", "target_text": "The converted voice follows the source content while adopting the requested style.", "source_audio": "resources/vevo2_official_gallery/style_whisper2normal_input.wav", "style_ref": "resources/vevo2_official_gallery/style_whisper2normal_ref.wav", "style_ref_text": "This reference carries the converted vocal style.", "use_pitch_shift": false, "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 } ] }, { "id": "vevo2_s2s_routes", "coverage": "VeVo2 speech-to-speech editing route", "family": "vevo2", "model": "models/Vevo2", "task": "s2s", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "editing", "route": "editing", "target_text": "Please replace the spoken phrase with this edited sentence.", "source_audio": "resources/vevo2_official_gallery/editing_child_gen.wav", "style_ref_text": "This reference preserves the original speaker style.", "use_pitch_shift": false, "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 } ] }, { "id": "vevo2_svc_routes", "coverage": "VeVo2 singing voice conversion task routes including style conversion and melody control", "family": "vevo2", "model": "models/Vevo2", "task": "svc", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "style_preserved_svc", "route": "style_preserved_svc", "source_audio": "resources/vevo2_official_gallery/mj.wav", "target_voice": "resources/vevo2_official_gallery/adele_raw.wav", "use_pitch_shift": false, "seed": 1234, "num_inference_steps": 32 }, { "id": "style_converted_svc", "route": "style_converted_svc", "target_text": "A steady phrase follows the converted singing style.", "source_audio": "resources/vevo2_official_gallery/style_singing_input.wav", "style_ref": "resources/vevo2_official_gallery/style_singing_ref.wav", "style_ref_text": "This reference carries the target singing style.", "use_pitch_shift": false, "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 }, { "id": "singing_style_conversion", "route": "singing_style_conversion", "target_text": "The same singer changes style while keeping the melody stable.", "source_audio": "resources/vevo2_official_gallery/style_singing_input.wav", "style_ref": "resources/vevo2_official_gallery/style_singing_ref.wav", "style_ref_text": "This reference carries the target singing style.", "use_pitch_shift": false, "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 }, { "id": "humming_to_singing", "route": "humming_to_singing", "target_text": "The melody becomes a sung line with clear words.", "prosody_ref": "resources/vevo2_official_gallery/humming_melody_ref.wav", "style_ref": "resources/vevo2_official_gallery/taiyizhenren.wav", "style_ref_text": "This reference carries the target singing voice.", "target_voice": "resources/vevo2_official_gallery/taiyizhenren.wav", "use_pitch_shift": false, "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 }, { "id": "instrument_to_singing", "route": "instrument_to_singing", "target_text": "The piano melody becomes a sung line with matching phrasing.", "prosody_ref": "resources/vevo2_official_gallery/piano_melody_ref.wav", "style_ref": "resources/vevo2_official_gallery/taiyizhenren.wav", "style_ref_text": "This reference carries the target singing voice.", "target_voice": "resources/vevo2_official_gallery/taiyizhenren.wav", "use_pitch_shift": false, "seed": 1234, "temperature": 1.0, "top_k": 25, "top_p": 0.8, "num_inference_steps": 32 } ] }, { "id": "kokoro_near_cap_reuse", "coverage": "Kokoro long-short-long request sizes, frontend, predictor, decoder graph reuse", "family": "kokoro_tts", "model": "models/kokoro-82m-v1_0-ggml", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "near_cap", "text": "At the museum dinner, the last plates were cleared away before a game began, and the final play was a strange word that everyone debated with good humor for several minutes before agreeing to write it down.", "language": "a", "seed": 1234 }, { "id": "short", "text": "Silver carts wait by the pier.", "language": "a", "seed": 1234 }, { "id": "near_cap_again", "text": "Morning barges drifted past the bridge as the clerk reread the notice, tucked the blue envelope into her coat, and described the quiet station in a calm clear voice.", "language": "a", "seed": 1234 } ] }, { "id": "omnivoice_clone_long", "coverage": "OmniVoice reference clone path, generator chunking, audio tokenizer conditioning", "family": "omnivoice", "model": "models/OmniVoice", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "clone_long", "text": "OmniVoice should keep its reference speaker stable while generating a longer paragraph that exercises prompt building and audio tokenizer reuse.", "language": "en", "voice_ref": "resources/a.wav", "reference_text": "This little work was finished in the year eighteen o three, and intended for immediate publication.", "options": { "audio_chunk_duration": 15.0 }, "seed": 1234, "guidance_scale": 2.0, "num_inference_steps": 16 } ] }, { "id": "omnivoice_voice_design", "coverage": "OmniVoice instruction-only voice design path without reference-audio defaults", "family": "omnivoice", "model": "models/OmniVoice", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "design", "text": "This second request uses an instruction instead of a reference voice.", "language": "en", "instruct": "female, low pitch, british accent", "seed": 1234, "guidance_scale": 2.0, "num_inference_steps": 16 } ] }, { "id": "pocket_tts_long_short_long", "coverage": "PocketTTS text conditioner, voice conditioner, flow LM, Mimi decoder chunk/reuse path", "family": "pocket_tts", "model": "models/pocket-tts", "task": "tts", "mode": "offline", "session_options": { "language": "english" }, "outputs": [ "audio" ], "requests": [ { "id": "long", "text": "The museum guide paused at the bronze clock so the children could hear its careful chime echo through the hall while visitors moved toward the river doors.", "voice_ref": "resources/a.wav", "seed": 1234, "max_tokens": 160 }, { "id": "short", "text": "Rain taps softly on the glass roof.", "voice_ref": "resources/a.wav", "seed": 1234, "max_tokens": 96 }, { "id": "long_again", "text": "Another longer request checks that PocketTTS keeps its session state stable after a shorter sentence and does not rebuild more than needed.", "voice_ref": "resources/a.wav", "seed": 1234, "max_tokens": 160 } ] }, { "id": "moss_tts_nano_voice_clone_long", "coverage": "MOSS-TTS-Nano voice clone, text/audio token generation, MOSS-Audio-Tokenizer-Nano decode", "family": "moss_tts_nano", "model": "models/MOSS-TTS-Nano-100M", "task": "tts", "mode": "offline", "outputs": [ "audio" ], "requests": [ { "id": "long", "text": "The studio lights clicked on before the first phrase began, and the cloned voice should stay steady through this medium length request.", "voice_ref": "resources/a.wav", "reference_text": "This little work was finished in the year eighteen o three, and intended for immediate publication.", "seed": 1234, "max_tokens": 120, "do_sample": false } ] }, { "id": "qwen3_asr_offline", "coverage": "Qwen3 ASR offline audio encoder and thinker decode", "family": "qwen3_asr", "model": "models/Qwen3-ASR-0.6B", "task": "asr", "mode": "offline", "outputs": [ "text" ], "requests": [ { "id": "sample", "audio": "resources/sample.wav", "language": "English", "text": "Transcribe the speech exactly. Preserve numeric values." } ] }, { "id": "parakeet_tdt_offline_long", "coverage": "Parakeet offline pre-encode, long-context encoder, TDT decoder", "family": "parakeet_tdt", "model": "models/parakeet-tdt-0.6b-v3", "task": "asr", "mode": "offline", "session_options": { "parakeet_tdt.offline_mode": "long_form" }, "outputs": [ "text" ], "requests": [ { "id": "sample", "audio": "resources/sample_16k.wav" } ] }, { "id": "parakeet_tdt_streaming", "coverage": "Parakeet streaming chunk ingestion, buffered context, streaming finalize", "family": "parakeet_tdt", "model": "models/parakeet-tdt-0.6b-v3", "task": "asr", "mode": "streaming", "session_options": { "parakeet_tdt.audio_chunk_duration_sec": "2", "parakeet_tdt.left_context_sec": "10", "parakeet_tdt.right_context_sec": "2" }, "outputs": [ "text" ], "chunk_size": 16000, "requests": [ { "id": "stream", "audio": "resources/sample_16k.wav" } ] }, { "id": "qwen3_forced_aligner_words", "coverage": "Qwen3 forced aligner transcript conditioning and word timestamp output", "family": "qwen3_forced_aligner", "model": "models/Qwen3-ForcedAligner-0.6B", "task": "align", "mode": "offline", "outputs": [ "words" ], "requests": [ { "id": "align", "audio": "resources/sample.wav", "language": "English", "text": "Some call me nature. Others call me Mother Nature. I've been here for over 4.5 billion years. Twenty-two thousand five hundred times longer than you." } ] }, { "id": "citrinet_asr_offline", "coverage": "Citrinet ASR frontend and CTC decode", "family": "citrinet_asr", "model": "models/citrinet", "task": "asr", "mode": "offline", "outputs": [ "text" ], "requests": [ { "id": "sample", "audio": "resources/a.wav" } ] }, { "id": "silero_vad_streaming", "coverage": "Silero VAD streaming chunk state and final speech segments", "family": "silero_vad", "model": "assets/framework/models/silero_vad", "task": "vad", "mode": "streaming", "outputs": [ "segments" ], "chunk_size": 512, "requests": [ { "id": "stream", "audio": "resources/sample_16k.wav" } ] }, { "id": "marblenet_vad_offline", "coverage": "MarbleNet VAD offline frontend and segment postprocess", "family": "marblenet_vad", "model": "models/marblenet_vad", "task": "vad", "mode": "offline", "outputs": [ "segments" ], "requests": [ { "id": "sample", "audio": "resources/sample_16k.wav" } ] }, { "id": "sortformer_diar_offline", "coverage": "Sortformer diarization frontend, encoder graph, speaker-turn postprocess", "family": "sortformer_diar", "model": "models/diar_sortformer_4spk-v1", "task": "diar", "mode": "offline", "outputs": [ "turns" ], "requests": [ { "id": "sample", "audio": "resources/sample_16k.wav" } ] }, { "id": "htdemucs_long_source_separation", "coverage": "HTDemucs long input chunking and multi-stem output", "family": "htdemucs", "model": "models/htdemucs", "task": "sep", "mode": "offline", "outputs": [ "named_audio" ], "requests": [ { "id": "long_180s", "audio": "resources/mel_roformer_test_180s.wav" } ] }, { "id": "mel_band_roformer_vocal", "coverage": "Mel-band RoFormer source separation graph and mixture-minus-vocals output", "family": "mel_band_roformer", "model": "models/mel-roformer-mlx", "task": "sep", "mode": "offline", "outputs": [ "named_audio" ], "requests": [ { "id": "vocal", "audio": "resources/mel_roformer_test_15s.wav" } ] } ], "audit_gaps": [ { "family": "band_split_roformer", "model": "models/roformer-models/bs_roformer/multistem", "reason": "audiocpp_cli --inspect rejects this reference root; only vocals_viperx is currently detectable under the band_split_roformer checkpoint family" }, { "family": "band_split_roformer", "model": "models/roformer-models/bs_roformer/vocals_viperx", "reason": "native audiocpp_cli inference runtime is not implemented yet for the band_split_roformer checkpoint family" } ] }