Text-to-speech compared
Text-to-speech is the voice a phone agent speaks with. What separates the options for a live call is time to first audio, price per character and whether a British voice is on offer, not just how natural the demo sounds.
Too few independent figures to compare these side by side yet. List prices are on the second tab.
- Single figures
- 1 of 18Each from a different test, not comparable
- Vendor claims
- 9 of 18Publish their own figure
- Lowest price
- $0.0128Inworld Realtime TTS-2 Flash
- Highest price
- $0.05Rime Coda v2
18 compared, read 29 October 2026 or later. Click a bar to open its row, a column heading to sort, ↗ for the source. Tick Compare on up to four rows.
Showing 18 of 18
Too few independent figures to compare yet
Too few to compareNo independent source has measured enough of these 18 the same way to chart them side by side. Below: the single figures there are, who measures them elsewhere, and 9 vendor claims, folded and unranked. Price is on the next tab.
Measured by others, one at a time
Each from a different test, so they cannot be compared with each other.
- Hexgrad Kokoro-82M 4.51, predicted naturalness UTMOS22 on LibriSpeech English, out of 5, higher is better. T5Gemma-TTS technical report (arXiv 2604.01760). arxiv.org
Measured elsewhere
Independent leaderboards whose terms do not let us republish their figures, so we link to them instead.
- Artificial Analysis Text to Speech TTS quality via its Speech Arena (blind votes), latency and price
- TTS Arena V2 (blind listening votes) ELO rating from blind head-to-head votes on which voice sounds more human
What RingTru will measure
Time to first audio over a phone line, and how often a listener mishears what the voice said.
What each vendor says about its speed (9, unranked)
Each vendor’s own words, in name order. They time different things, so they are not charted or ranked.
- Azure Speech neural voices Less than 300 ms learn.microsoft.com
- Canopy Labs Orpheus TTS ~200ms streaming, reducible to ~100ms with input streaming github.com
- Cartesia Sonic 3.6 first audio in under 90ms cartesia.ai
- ElevenLabs Eleven v4 Turbo ~100ms median inference elevenlabs.io
- ElevenLabs Flash v2.5 ~75ms elevenlabs.io
- Hume AI Octave 2 ~100ms (Octave 2), ~200ms (Octave 1), not including network dev.hume.ai
- Resemble AI Chatterbox sub 200ms for Resemble's hosted service (not the open models) github.com
- Rime Coda v2 sub-100 ms median (P50) time to first audio at 1 concurrent request (vendor benchmark) rime.ai
- Rime Mist v3 37 ms median (P50), 56 ms P90 time to first audio rime.ai
- Price, lowestInworld Realtime TTS-2 Flash $0.0128
Bar length is the figure itself, so the shortest bar is best in every column.
| US$ a minute | Checks✓ yes, ✕ no, ~ partly | Measured by otherssingle figures | |
|---|---|---|---|
| ElevenLabs Flash v2.5 | $0.034 | Not published | None found |
| ElevenLabs Eleven v4 Turbo | $0.034 | Not published | None found |
| Cartesia Sonic 3.6 | $0.0425 | British voices | None found |
| Deepgram Aura-2 | $0.0255 | British voices | None found |
| OpenAI gpt-4o-mini-tts | Not comparable | British voices | None found |
| Rime Mist v3 | $0.03 | British voices | None found |
| Rime Coda v2 | $0.05 | British voices | None found |
| Hume AI Octave 2 (preview) | Not published | Not published | None found |
| Inworld Realtime TTS-2 Flash | $0.0128Lowest | Not published | None found |
| Inworld Realtime TTS-2 | $0.0213 | Not published | None found |
| Microsoft Azure Azure Speech neural and HD voices | $0.0128Lowest | British voices | None found |
| Google Cloud Chirp 3: HD voices | $0.0255 | British voices | None found |
| Amazon Web Services Amazon Polly (generative and neural voices) | $0.0255 | British voices | None found |
| Resemble AI Resemble text-to-speech API | Not published | Not published | None found |
| Hexgrad (community) Kokoro-82MOpen weights | Not published | British voices | |
| Canopy Labs Orpheus TTSOpen weights | Not published | British voices | None found |
| Resemble AI (open source) Chatterbox (Turbo, Multilingual, Nano)Open weights | Not published | Not published | None found |
| Sesame CSM-1BOpen weights | Not published | Not published | None found |
Vendor claims, data regions and compliance are in each row’s Details, in the vendor’s own words. Claims are never charted or ranked. RingTru has measured none of these yet. Add up a whole call in the calculator.
How to read these numbers
- Price units differ
- Some vendors charge per character, some per minute of audio and some sell monthly credits. The calculator converts characters to minutes at 150 words a minute, about 850 characters, and says so.
- Time to first audio
- The time from sending text to receiving the first sound. It is the part of text-to-speech a caller notices. Vendors usually measure it close to their own servers, so a real call adds network time.
- British voices
- A British voice is more than an accent setting. Check how it reads UK postcodes, phone numbers and place names before choosing it.
Go deeper
Your agent, tested on real calls.
- test calls
- 506
- passed
- 96.8%
- median reply
- 0.94 s
Holds the new day when a caller changes their mind.