Best Controlled Voice Text to Speech (TTS) Models
Compare quality, speed, and price of speech generation models with every model speaking the same cloned voices, isolating model quality from voice selection.
For further details, see our methodology page.
Quality
Controlled Voice Arena Preference Rank by Language
Arena Elo ranking by language · Lower rank is better
Models with results for at least 6 languages are ranked first by their average normalized rank: (rank - 1) / (models ranked in that language - 1). Only languages with at least two ranked models count toward coverage and the average.
–: No preference rank yet. Samples may not have been generated, or the model may not currently provide that language.
Controlled Voice Arena Preference English (US & UK) vs. Price
Arena Elo: average Elo rating of the model · Price: USD per 1M characters of text
Most attractive quadrant
Pareto line
Pricing
Price
Price: USD per 1M characters of text · Lower is better
Latency
Time to First Audio
Average time to first audio across all requests and prompt lengths (milliseconds) · Lower is better
Streaming
Non-streaming (complete response)
Latency data available for 27 of 31 selected models
Last 7 days: 6 measurement days: 23 models
Last 7 days: 1 measurement day: Eleven v4 Turbo, ElevenLabs and Eleven v4, ElevenLabs
Last 7 days: 2 measurement days: v3 Conversational, ElevenLabs and Eleven v3, ElevenLabs
Throughput
Characters Per Second
Characters processed per second: # of characters per second of generation time · Higher is better