Voiceover for a phone voice agent
Which text-to-speech service speaks fastest and still sounds natural for a customer-service voice agent, for the least money?
JSONAs of 5 November 2026, no option is recommended. Quality has not been measured for any of the 25 options on this board, and cost is not a recommendation.
- Options
- 25
- Measured
- 0
- Cost confirmed
- 14
- Not confirmed
- 9
- No price
- 2
The job · requires commercial rights, English
Run a voice agent that answers 1,500 English calls a month. Latency matters because the caller waits for the first sound. Required: commercial rights; English.
Nobody is ranked. Not plotted. Quality has not been measured, so the list below is cost only and not an order of merit.
Cost for this job
Cheapest qualifying plan per month, in USD per month, cheapest first within each group. Not in order of merit; says nothing about quality.
Workload and assumptions
Workload: 3,000,000 characters a month. 1,500 calls x 2 minutes of agent speech = 3,000 minutes = 3,000,000 characters. Callers speak the other half of the call. App plans (Fish Audio app, Murf Studio) are left out because they have no API. 1,000 characters per finished minute of audio (assumption: ElevenLabs lists about $0.08 per minute at $0.08 per 1,000 characters; Inworld says about 1,000; Cartesia says 750 to 800).
| Option | Cost | Plan | Price read | Source | Notes |
|---|---|---|---|---|---|
| Cost confirmed against the requirements 14 | |||||
Polly StandardAmazon | $12.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Falcon 2.1Murf | $30.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Aura-1Deepgram | $45.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Polly NeuralAmazon | $48.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Aura-2Deepgram | $90.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Gen2Murf | $90.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Polly GenerativeAmazon | $90.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Eleven FlashElevenLabs | $120.00 | creator$22.00 a month, subscription | 2 days old | 1 note
| |
Eleven v4 TurboElevenLabs | $120.00 | creator$22.00 a month, subscription | 2 days old | 1 note
| |
Sonic 3.6Cartesia | $127.75 | startup$49.00 a month, subscription | 2 days old | 1 note
| |
Speech 2.8 TurboMiniMax | $180.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Eleven v4ElevenLabs | $239.96 | pro$99.00 a month, subscription | 2 days old | 1 note
| |
Speech 2.8 HDMiniMax | $300.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
Polly Long-FormAmazon | $300.00 | pay-as-you-go$0.00 a month, usage | 2 days old | ||
| Cost not confirmed: a requirement is unknown 9 | |||||
Standard voicesGoogle Cloud | No charge listednot confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 1 note
| |
WaveNetGoogle Cloud | No charge listednot confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 1 note
| |
TTS-2 FlashInworld | $30.00not confirmed | creator$25.00 a month, subscription | 2 days old | 2 notes
| |
Neural2Google Cloud | $32.00not confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 1 note
| |
Neural voicesMicrosoft Azure | $45.00not confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 2 notes
| |
Chirp 3 HDGoogle Cloud | $60.00not confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 1 note
| |
TTS-2Inworld | $60.00not confirmed | creator$25.00 a month, subscription | 2 days old | 2 notes
| |
Flux TTSDeepgram | $135.00not confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 1 note
| |
Studio voicesGoogle Cloud | $320.00not confirmed | pay-as-you-go$0.00 a month, usage | 2 days old | 1 note
| |
| No cost available 2 | |||||
S2.1 ProFish Audio | Unknown | No plan | Not read | No source | 1 note
|
gpt-4o-mini-ttsOpenAI | Unknown | No plan | Not read | No source | 1 note
|
What evidence would settle it
The three conditions and what is missing
Vendor claims about their own quality count for nothing in the score. A rating needs a listening test with raw results, or a licensed source that allows reuse.
- Measure quality for Neural voices, Sonic 3.6, Aura-1, Aura-2, Flux TTS, Eleven Flash and 19 more: Time to first audio (ms, median, streaming API called from Europe); Naturalness in blind listening (% of blind pairwise votes won against the board's reference voice); Pronunciation errors (mispronounced words per 1,000 words). At least one auditable or documented measurement per option.
- Name at least two options that are measured. Today 0 are.
The three conditions for naming a best option
- The leader's lower bound must exceed every other option's upper bound by at least 3 points. Not assessed.
- Every critical metric must come from two independent origins, or from one auditable test of our own. Not assessed.
- The price must be verified within 30 days. Not assessed.
Sources and limits
11 sources, the method notes and this board as JSON
No quality score is published on this board. Prices are facts read from each vendor's page and link back to it.
- Amazon Polly pricing prices only, as facts with a link; checked
- Azure Speech pricing prices only, as facts with a link; checked
- Cartesia pricing prices only, as facts with a link; checked
- Deepgram pricing prices only, as facts with a link; checked
- ElevenLabs API pricing prices only, as facts with a link; checked
- Fish Audio API and app pricing prices only, as facts with a link; checked
- Google Cloud Text-to-Speech pricing prices only, as facts with a link; checked
- Inworld pricing prices only, as facts with a link; checked
- MiniMax Speech pay-as-you-go pricing prices only, as facts with a link; checked
- Murf pricing prices only, as facts with a link; checked
- OpenAI gpt-4o-mini-tts model page prices only, as facts with a link; checked
Notes on this board. No quality evidence is published yet: the rights register has no usable source for voice quality (see the Sources page). The board shows cost only and ranks nobody until measured or licensed evidence arrives. Quality metrics and anchors are editorial and unvalidated.
Scores are on a 0 to 100 scale against editorial anchors that are not yet validated. Prices are for the workload above and change without notice. Evidence older than 60 days is left out, and prices older than 30 days are flagged.
This board as JSON, with every input id, anchor and weight.