Voice AI Benchmarks By

Overview Speech-to-Text Text-to-Speech Speech-to-Speech Methodology

FEATURED MODELS

Nova 3Speech-to-text Eleven v3 ConversationalText-to-speech Gemini LiveSpeech-to-speech

Explore all models→

FEATURED PROVIDERS

DeepgramSTT and TTS ElevenLabsSTT and TTS OpenAISTT, TTS and S2S

Explore all providers→

FEATURED BENCHMARKS

Word Error RateAccuracy · STT and TTS Time to First AudioLatency · TTS Voice-to-Voice LatencyLatency · S2S

Explore all benchmarks→

FEATURED DATASETS

WildASR CleanSTT · Clean baseline TTS Prompt SetTTS · Fixed prompts Multi-turn CallsS2S · Simulated calls

Explore all datasets→

LEADERBOARDS

Overview Speech-to-Text Text-to-Speech Speech-to-Speech Methodology

EXPLORE

Models

Nova 3Speech-to-text Eleven v3 ConversationalText-to-speech Gemini LiveSpeech-to-speech Explore all models→

Providers

DeepgramSTT and TTS ElevenLabsSTT and TTS OpenAISTT, TTS and S2S Explore all providers→

Benchmarks

Word Error RateAccuracy · STT and TTS Time to First AudioLatency · TTS Voice-to-Voice LatencyLatency · S2S Explore all benchmarks→

Datasets

WildASR CleanSTT · Clean baseline TTS Prompt SetTTS · Fixed prompts Multi-turn CallsS2S · Simulated calls Explore all datasets→

Theme

Interactive benchmark

Voice Arena leaderboard

Models ranked by Elo rating from blind A/B votes. The ± figure is the confidence interval — the fewer votes a model has, the wider it runs.

Loading the leaderboard…

Voice AI Benchmarks

Independent, continuously run comparisons of speech recognition, synthesis and native-audio models by Coval.

Leaderboards

Overview Speech-to-Text Text-to-Speech Speech-to-Speech Arena

Explore

Models Providers Benchmarks Datasets

Methodology

Benchmark methodology Open-source methodology Benchmark runner Coval on GitHub

Coval

Coval.ai Pricing Book a Demo

© 2026 Coval, Inc.

X LinkedIn Email Privacy Policy Your Privacy Choices