Amazon Polly vs Speakvora: a one-stream speed test
Our Phonon voices reached the first audible sound in 32 to 56 ms against 146 to 180 ms for Amazon Polly Neural. One stream, the same machine, raw data and audio below.
Tested November 9, 2026 on an Apple M2 MacBook (16.0 GB). One stream at a time, 10 runs per phrase and engine, four phrases. Our own test, not a third-party benchmark.
The result, one stream
| Voice | First audible sound, median per phrase | Slowest 5% (worst phrase) | First bytes arrive | Leading silence |
|---|---|---|---|---|
| Speakvora Phonon Delia | 38 to 56 ms | 65 ms | 27 to 44 ms | 11 to 13 ms |
| Speakvora Phonon Easton | 32 to 45 ms | 46 ms | 20 to 34 ms | 11 to 13 ms |
| Speakvora Pocket Cole | 104 to 154 ms | 171 ms | 90 to 121 ms | 12 to 24 ms |
| Amazon Polly Neural Joanna | 146 to 180 ms | 565 ms | 54 to 70 ms | 92 to 111 ms |
Time to the first audible sound is the first audio chunk plus any silence at the start of the clip, the way Coval's published TTS benchmark describes it, from our own implementation. Phonon voices were fastest here (32 to 56 ms), Pocket voices were next (104 to 154 ms) and Amazon Polly Neural took 146 to 180 ms, of which about 92 to 111 ms was silence at the start of its audio.
How we tested
- The Speakvora voice server ran on the same MacBook as the test script, so Speakvora rows have no internet hop. Amazon Polly was called from the same MacBook over the internet (AWS us-east-1), so Polly rows include that network path. This favors the local server, and that is why we publish the first-bytes column next to the headline.
- One stream at a time. Every phrase and engine was run in turn so the network and the machine affect all of them alike. Each engine was warmed up once first.
- Not tested: other Polly engines or voices, long text, other networks or regions, concurrent users. This page does not say how many streams anything can serve.
- The voice server on the Mac was also answering free-account traffic during the test.
Every phrase
| Phrase | Voice | First audible, ms (median) | Slowest 5%, ms | First bytes, ms | Leading silence, ms | Runs ok |
|---|---|---|---|---|---|---|
| Short | Speakvora Phonon Delia | 38 | 50 | 27 | 11 | 10/10 |
| Short | Speakvora Phonon Easton | 32 | 41 | 20 | 12 | 10/10 |
| Short | Speakvora Pocket Cole | 104 | 123 | 90 | 14 | 10/10 |
| Short | Amazon Polly Neural Joanna | 166 | 172 | 55 | 111 | 10/10 |
| Medium | Speakvora Phonon Delia | 48 | 65 | 36 | 12 | 10/10 |
| Medium | Speakvora Phonon Easton | 45 | 46 | 34 | 11 | 10/10 |
| Medium | Speakvora Pocket Cole | 154 | 171 | 121 | 24 | 10/10 |
| Medium | Amazon Polly Neural Joanna | 161 | 168 | 54 | 107 | 10/10 |
| Numbers | Speakvora Phonon Delia | 52 | 59 | 39 | 13 | 10/10 |
| Numbers | Speakvora Phonon Easton | 32 | 33 | 20 | 12 | 10/10 |
| Numbers | Speakvora Pocket Cole | 113 | 127 | 96 | 15 | 10/10 |
| Numbers | Amazon Polly Neural Joanna | 180 | 565 | 70 | 110 | 10/10 |
| Question | Speakvora Phonon Delia | 56 | 62 | 44 | 12 | 10/10 |
| Question | Speakvora Phonon Easton | 33 | 38 | 20 | 13 | 10/10 |
| Question | Speakvora Pocket Cole | 105 | 136 | 94 | 12 | 10/10 |
| Question | Amazon Polly Neural Joanna | 146 | 160 | 54 | 92 | 10/10 |
Listen
One recording per voice and sentence (the median run), converted to MP3. Hear the same words from each engine.
Short: “Your order is ready for pickup.”
- Speakvora Phonon Delia
- Speakvora Phonon Easton
- Speakvora Pocket Cole
- Amazon Polly Neural Joanna
Medium: “Thanks for calling. We received your request this morning. A technician has been assigned to your address.”
- Speakvora Phonon Delia
- Speakvora Phonon Easton
- Speakvora Pocket Cole
- Amazon Polly Neural Joanna
Numbers: “Your appointment is on Tuesday, November 3rd at 2:45 PM, and the total is $1,249.50.”
- Speakvora Phonon Delia
- Speakvora Phonon Easton
- Speakvora Pocket Cole
- Amazon Polly Neural Joanna
Question: “I can help with that. Can you confirm the phone number on the account?”
- Speakvora Phonon Delia
- Speakvora Phonon Easton
- Speakvora Pocket Cole
- Amazon Polly Neural Joanna
Check the data
Raw results (JSON) · Automatic transcription check (JSON) · Machine details (JSON)
The transcription check ran Whisper (base.en) on each recording and compared the words with the sentence. Phonon and Pocket matched the sentence exactly on all four phrases; Amazon Polly matched exactly on two, and on the other two the check wrote "pick up" for "pickup" and "3" for "3rd", which is a spelling difference in the checker, not garbled audio. It is a smoke test, not a ranking of voice quality.
More: Text to speech latency · the fastest text to speech API ranked · · switch from Amazon Polly
Questions
Is Speakvora faster than Amazon Polly?
In this one-stream test on November 9, 2026, our Phonon voices reached the first audible sound in 32 to 56 ms and Polly Neural in 146 to 180 ms (median per phrase, 10 runs each, same machine). The conditions matter: our voice server ran on the same Mac, and Polly was called over the internet. See the method and the limits below.
Why is Polly's number higher if its first bytes arrive quickly?
Polly's first bytes arrived in 54 to 70 ms, but its audio starts with about 92 to 111 ms of silence. We count the time to the first audible sound (the first chunk plus any leading silence, the way Coval's TTS benchmark describes it), so that silence is included.
How many streams did you test?
One at a time. This page makes no claim about how many simultaneous streams Speakvora or Polly can serve.
Can I check the data?
Yes. The raw results, the automatic transcription check and the machine details are linked below, and every sentence has audio from each voice so you can compare by ear.
What did you not test?
Other Polly engines and voices, long text, other regions or networks, concurrent load, and voice quality by listening panel. Speed is one factor; try your own text before you choose.
Part of our text to speech API guide: voices, price, speed and code samples.