Speech AI pricing report
What text-to-speech, speech-to-text and realtime voice APIs cost, worked out from the published list prices in the Voicedex catalog. The figures are recomputed from the data every time it changes; last sync 20 November 2026.
284 text-to-speech models440 speech-to-text models67 speech-to-speech models126 providersHow this is calculated
Key findings
- Text-to-speech list prices vary 19-fold: the middle 90% of priced models cost between $5.30 and $100 per 1M characters. The median is $30, and half cost between $15 and $45.
- Speech-to-text varies 13-fold: the middle 90% cost between $0.11 and $1.40 per hour of audio, with a median of $0.72.
- Streaming costs more. The median speech-to-text model that streams over WebSocket or WebRTC lists at $1.20 per hour of audio, against $0.35 for models that don't: 3.4× as much.
- Among the most-covered languages, language barely moves the median list price: text-to-speech is $28 per 1M characters for English and $30 across the next 9; speech-to-text is $0.36 and $0.32 per hour.
- Prices are hard to find. Only 33% of text-to-speech models publish a per-character price in USD; 29 publish in another currency only, and 89 hosted models have no public price at all.
- Realtime speech-to-speech is mostly sold by the token: 33 of 67 models, so only 15 can be compared per hour of audio.
Where prices fall
Each square is one model at its published list price, on a log scale. Lit squares stream over WebSocket or WebRTC; hover or tap one to see it. The table below gives the same distribution in numbers.
Speech-to-text and realtime $ per hour of audio, log scale 239 priced
Text-to-speech $ per 1M characters, log scale 94 priced
other models streams (WebSocket or WebRTC)All models by price
| Type | Priced models | 5th percentile | Lower quartile | Median | Upper quartile | 95th percentile |
|---|---|---|---|---|---|---|
| Text-to-speech USD per 1M characters | 93 | $5.30 | $15.00 | $30.00 | $45.00 | $100.00 |
| Speech-to-text USD per hour of audio | 222 | $0.11 | $0.26 | $0.72 | $1.20 | $1.40 |
| Speech-to-speech USD per hour of audio | 15 | – | $3.00 | $3.00 | $4.80 | – |
For the cheapest options in a language, see the language pages, e.g. Hindi text-to-speech or Spanish speech-to-text, sorted by price with a monthly cost calculator.
Streaming vs not streaming
Median list price of models that stream over WebSocket or WebRTC against those that don't (HTTP request and response, or batch files).
| Type | Streaming: median | Not streaming: median | Ratio |
|---|---|---|---|
| Text-to-speech USD per 1M characters | $30.00 65 models | $17.25 28 models | 1.7× |
| Speech-to-text USD per hour of audio | $1.20 131 models | $0.35 91 models | 3.4× |
Text-to-speech prices by language
The 10 languages with the most priced text-to-speech models. A model counts for every language it lists; models that support "many languages" without listing them aren't counted.
| Language | Priced models | Median | Middle half |
|---|---|---|---|
| English | 80 | $28.00 | $15.00 to $60.00 |
| German | 65 | $30.00 | $15.00 to $60.00 |
| Spanish | 65 | $30.00 | $15.00 to $65.00 |
| French | 65 | $30.00 | $15.00 to $65.00 |
| Portuguese | 60 | $30.00 | $15.00 to $61.25 |
| Chinese | 60 | $30.00 | $15.00 to $65.00 |
| Italian | 59 | $30.00 | $15.00 to $62.50 |
| Japanese | 58 | $30.00 | $15.00 to $65.00 |
| Korean | 57 | $30.00 | $15.00 to $65.00 |
| Russian | 54 | $30.00 | $15.00 to $65.00 |
USD per 1M characters.
Speech-to-text prices by language
The 10 languages with the most priced speech-to-text models. A model counts for every language it lists; models that support "many languages" without listing them aren't counted.
| Language | Priced models | Median | Middle half |
|---|---|---|---|
| English | 133 | $0.36 | $0.21 to $0.96 |
| Spanish | 86 | $0.35 | $0.15 to $0.96 |
| French | 82 | $0.32 | $0.15 to $0.94 |
| Portuguese | 76 | $0.32 | $0.15 to $0.89 |
| German | 75 | $0.32 | $0.15 to $0.73 |
| Japanese | 75 | $0.32 | $0.15 to $0.73 |
| Chinese | 74 | $0.31 | $0.14 to $0.80 |
| Italian | 72 | $0.32 | $0.15 to $0.67 |
| Korean | 70 | $0.29 | $0.14 to $0.60 |
| Hindi | 68 | $0.31 | $0.15 to $0.51 |
USD per hour of audio.
How many models publish a price
"Comparable" means a USD price in the unit the catalog normalizes to. The rest are counted, not compared: we never convert currencies or guess a per-character rate from tokens.
| Type | Models | Comparable USD price | USD, other unit | Other currency only | No public price | Open weights, self-hosted |
|---|---|---|---|---|---|---|
| Text-to-speech | 284 | 93 33% | 25 | 29 | 89 | 47 |
| Speech-to-text | 440 | 222 50% | 10 | 55 | 137 | 16 |
| Speech-to-speech | 67 | 15 22% | 33 | 11 | 7 | 1 |
Other currencies: text-to-speech: CNY 16, KRW 5, INR 3, UZS 1, JPY 1, GBP 1, MNT 1, EUR 1; speech-to-text: CNY 17, JPY 17, KRW 9, EUR 6, INR 3, UZS 1, MNT 1, RUB 1; speech-to-speech: CNY 11.
How this is calculated
- Source. Each price is read from the provider's own pricing page or API and stored with its source link and the date it was checked; every model page shows both. The catalog is versioned in git, so every price change has a history.
- Which price. The published list price anyone can pay without a subscription commitment. Plan-gated rates, volume discounts and free tiers are recorded on each model page but not used here.
- What's counted. Active models from providers that are still selling, with each
-latestalias counted once. Prices are normalized to USD per 1M characters for text-to-speech and USD per hour of audio for speech-to-text and speech-to-speech; prices in other currencies or units are counted in the last table but not compared. - Ranges, not extremes. Spreads use the 5th to 95th percentile, so one unusual price at either end can't distort them. Single cheapest models change often and some depend on conditions, so they are on the price-sorted pages, with notes, rather than here.
- Streaming means the model accepts or returns audio over WebSocket or WebRTC.
Download every model as CSVFull catalog as JSONWhat changedSpot an error? Report it