ListenSpeedScoresCapabilities🗳 Vote ↗

Scores

Objective scores over the same 5 prompts. UTMOS = predicted naturalness (higher better); WER = ASR word-error rate vs the intended text — a failure-detector, not a fine ranking (lower better); SIM = speaker similarity to the cloned reference (the house reference, higher better); Health = deterministic defect triage of the published clip (⚠ flags long internal silence / clipping / dead audio — a "go listen" cue, not a score). Switch Default / Cloning below; click any header to re-sort. Each score is the mean over the exact clips shown on Listen. Human votes are the preference ground truth; these objective metrics are backstops.
Voice:
Default voiceCloning

Default voice · naturalness + intelligibility

Top 15 · metric:
ModelSizeReleasedUTMOS ↑WER ↓Health
Audio8 TTS 0.6B (compiled)601MAug 20264.1230.040✓
Audio8 TTS Preview 0.1B170MAug 20264.1220.037✓
Audio8 TTS Preview 0.6B601MJul 20264.2260.049✓
AuroraTTS-100K95.6KOct 20263.1110.116✓
Chatterbox1.2BApr 20254.4230.061✓
Chatterbox Turbo744MDec 20254.2740.074✓
Coqui XTTS-v2750MOct 20233.9360.054✓
Dia 1.6B-06261.6BJun 20252.3870.317✓
dots.tts (soar)2BJun 20263.2270.059✓
DramaBox3.3BApr 20264.3040.093✓
Higgs Audio v3 TTS4BJun 20264.3720.065✓
Inflect-Micro v29.36MJul 20264.3890.030✓
Inflect-Nano v23.96MJul 20264.3540.059✓
KittenTTS 21.7BSep 20264.2890.035✓
KittenTTS 2 (C++ CPU)1.7BSep 20264.2710.029✓
KittenTTS Nano 0.1<100MAug 20253.6650.093✓
KittenTTS Nano 0.815MFeb 20264.1670.044✓
Kokoro82MDec 20244.1770.054✓
LFM2.5-Audio 1.5B1.5BDec 20254.3830.066✓
LongCat-AudioDiT 1B1.42BMar 20264.0530.093✓
LongCat-AudioDiT 3.5B3.83BMar 20264.3410.080✓
Magpie-TTS357MDec 20254.1990.087✓
Mars5-TTS1.2BJun 20243.5400.361✓
Maya13BOct 20254.4870.066✓
MeloTTS~52MFeb 20243.4950.059✓
Ming-3.0-TTS 0.5B1.48BOct 20263.7350.034✓
Ming-3.0-TTS tiny8.98B (MoE, 1.3B active)Oct 20264.2920.041✓
MiraTTS0.5BDec 20253.8030.100✓
Miso TTS 8B8.2BMay 20264.2320.586✓
OmniVoice~1BMar 20264.1040.040✓
Orpheus-TTS 3B3.3BMar 20254.0050.102✓
OuteTTS 1.0 1B1BApr 20254.3860.070✓
Parler-TTS Mini v1878MJun 20243.7570.149⚠ gap (3)
Phonon 31m41.1M (30.8M LM + Mimi)Oct 20264.4240.073✓
Phonon 40m50.5M (40.2M LM + Mimi)Oct 20264.2520.079✓
Piper~25MBJan 20233.8770.055✓
Pocket-TTS v4100MJan 20263.9910.070✓
Qwen3-TTS 0.6B CustomVoice0.6BJan 20264.2680.110✓
sanoTTS Amy1.46MJul 20263.8210.067✓
sanoTTS Heart-Nano294KSep 20262.1540.019✓
Scylla's Band~103MJul 20264.3410.065✓
Sesame CSM-1B1BMar 20254.1520.114✓
Soprano 1.1 80M80MJan 20264.1160.059✓
SteerTTS40.8MOct 20264.1780.065✓
SteerTTS (no refiner)23.1MOct 20263.8780.065✓
StyleTTS 2~148MJun 20234.2590.158✓
Supertonic 399MMay 20264.0800.054✓
SupraTTS 0.1 Beta43.5MSep 20263.3230.084✓
Vaniq-Edge8.91MAug 20264.2290.115✓
VibeVoice Realtime 0.5B0.5BDec 20254.0430.148✓
VoxCPM2 2B2BApr 20263.4800.023✓
Voxtral 4B TTS4BNov 20253.6920.081✓
Scored over the 5 bench prompts (thin — WER is a failure-detector, not a fine ranking). Checkpoints: UTMOS utmos22_strong (SpeechMOS), SIM canonical UniSpeech-SAT wavlm_large_finetune, WER Whisper-large-v3. Method follows seed-tts-eval. Human votes are the preference ground truth; these are objective backstops.