Amazon Polly's Standard voices and Google's WaveNet and Standard voices each cost $4 per million characters, so a team reading appointment reminders aloud gets a quote that looks like a rounding error. The catch is that no voice from that tier was in the realism results, which means the $4 price has no score next to it, and a buyer comparing the two for a live call ends up pricing the newer tiers instead.
Those newer tiers cost more and work differently. Google sells Chirp 3: HD voices by the character and its Gemini text to speech models by the token, and Amazon sells Generative and Long-Form voices by the character at $30 and $100 per million. Bland Speech sits beside the two big clouds as the third option, a voice built for phone calls.
Which is better, Google text to speech or Amazon Polly?
As of November 2026, Google text to speech is the better fit for a team that wants expressive, tag-directed voices and a score in the realism results, and Amazon Polly is the better fit for a team that wants a flat per-character price with the largest free allowance. Google's Gemini 2.5 Pro TTS ranked 5th in Design Arena's Audio Realism Bench results of 4 August 2026, and Amazon Polly wasn't in those results.
What these rows mean
- Lowest published rate: The lowest published rate for a voice each company still sells. Neither the $4 Google voices nor the $4 Polly voices were in the realism results.
- Newer tier: Google bills Gemini-TTS in tokens, with 25 audio tokens per second of audio, so a minute of speech is 1,500 output tokens. Amazon bills every engine in characters.
- Streaming: Bidirectional streaming means text goes in and audio comes out on one open connection, which suits a language model writing a reply a few words at a time.
- Phone audio: Mu-law and a-law are compressed 8-bit formats used on phone networks, mu-law in North America and Japan.
- Realism result: The Elo score from Design Arena's Audio Realism Bench results of 4 August 2026, where listeners pick which of two clips sounds more like a real person.
What does Google text to speech do well, and where does it fall short?
Google text to speech does well on language range and on directing a performance, and it falls short on where its newest models live. Google says its voices span 75+ languages and variants with 380+ voices, and its Gemini-TTS models take inline tags in square brackets, such as [sigh], [laughing], [whispering], and [short pause], plus a style prompt that Google calls "the primary driver of the overall emotional tone and delivery."
Two catches come from Google's own wording. Google warns that its vocalized-adjective tags, such as [scared] and [curious], are spoken aloud, "likely an undesired side effect for most use cases." And Gemini 3.8 Flash TTS and Flash-Lite TTS are generally available "through the Agent Platform API and Agent Studio" but "aren't available through the Cloud Text-to-Speech API", so a team using Google's Cloud TTS API reaches Gemini 3.1 Flash TTS in Preview and the 2.5 models, while the 3.8 models sit on a second platform.
Google pricing
Google bills older voices by the character: WaveNet and Standard at $4 per million, Neural2 and Polyglot at $16, and Studio at $160. Chirp 3: HD costs $30 per million characters, and the first 1 million characters each month are free. Instant custom voice costs $60 per million characters. Gemini-TTS bills tokens: Gemini 2.5 Flash TTS is $0.50 per million text tokens in and $10.00 per million audio tokens out, and Gemini 2.5 Pro TTS and Gemini 3.1 Flash TTS are $1.00 in and $20.00 out. At 25 audio tokens per second, a minute of Gemini 2.5 Pro TTS speech comes to $0.03 in output tokens, plus the text going in. New Google Cloud customers get up to $300 in free credits.
Google controls and cloning
Chirp 3: HD has a speaking rate from 0.25 to 2.0 and pause tags, with no emotion control, and SSML isn't supported on its streaming requests. Google's Chirp 3 Instant Custom Voice carries a caution that "You can no longer request access", and Google points to voice replication in Gemini 3.8 TTS, which is in Preview. Google describes its streaming as "ultra-low-latency".
Pros
- 75+ languages and variants, 380+ voices
- Bracket tags and a style prompt on Gemini-TTS
- Chirp 3: HD with 1 million characters free each month
- Text-to-Speech on Google Cloud's HIPAA BAA product list
Cons
- Instant custom voice closed to new access requests
- Gemini 3.8 models sit outside the Cloud TTS API
- Some Gemini-TTS tags are spoken aloud
Google text to speech is a poor fit for a team that wants a cloned brand voice on a generally available model inside the Cloud TTS API.
What does Amazon Polly do well, and where does it fall short?
Amazon Polly does well on price and AWS fit, and it falls short on expressiveness and cloning. Polly's four engines are standard, neural, generative, and long-form, priced per million characters at $4, $16, $30, and $100, and AWS gives Standard voices a free tier of 5 million characters a month. Cached audio can be replayed at no extra cost, and new AWS customers get up to $200 in free credits. AWS calls the Generative engine "the largest Amazon Polly TTS model to-date", and Polly supports bidirectional streaming over HTTP/2 on that engine.
The catches come from AWS's own wording. AWS says generative voices can vary: "any updates to the training data and the model could result in slight variations to the way the voices sound." AWS also says its stop mechanism for hallucinated speech doesn't remove the risk, and speech marks aren't available for generative voices. SSML is a subset of the W3C standard with controls such as pauses, pace, volume, whispering, and breathing sounds, and the <audio>, <lexicon>, <lookup>, and <voice> tags aren't supported.
Amazon Polly limits and cloning
A single Polly request takes up to 3,000 billed characters. Cloning isn't self-serve: AWS says a team interested in a Brand Voice should "reach out to your AWS Account Manager."
On compliance, AWS calls Polly a HIPAA Eligible Service under the AWS BAA and in scope for SOC 1, 2, and 3, and says Polly is certified for use with regulated workloads for HIPAA and PCI DSS. Polly's output formats include mu-law and a-law at 8,000 Hz. AWS says Polly can "stream audio to your users in near real time."
Pros
- Standard voices at $4 per million characters, 5 million free each month
- Four engines at four price points
- HIPAA Eligible under the AWS BAA, in scope for SOC 1, 2, and 3
- Mu-law and a-law at 8,000 Hz
Cons
- Brand Voice cloning only through an AWS Account Manager
- Generative voices can shift between model updates
- A subset of SSML, without the audio and voice tags
Amazon Polly is a poor fit for a team that needs a cloned brand voice on a self-serve plan.
Which sounds more human, Google text to speech or Amazon Polly?
Google's Gemini models ranked in Design Arena's Audio Realism Bench results of 4 August 2026, and Amazon Polly wasn't in those results, so only one side of this pair has a score. Gemini 2.5 Pro TTS ranked 5th of 13 AI voices at 1109, Gemini 3.1 Flash TTS ranked 8th at 1037, and Gemini 2.5 Flash TTS ranked 10th at 985. Elo is a rating that rises each time listeners pick a voice as more human.
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
Ranks are among 13 AI voices, and Google's own voice is on the list. Real human recordings scored 1500, at the right edge of this scale. Bars start at 600, not zero. Source: Design Arena's published results.
Google's Chirp 3: HD voices and Gemini 3.8 models weren't in the results either, and neither were Polly's Standard, Neural, Generative, or Long-Form voices. In those same results Bland Speech v3 ranked 1st at 1365, and the human recording scored 1500. Design Arena publishes its method in an Audio Realism Bench article.
Where does Bland Speech fit next to Google text to speech and Amazon Polly?
Bland Speech fits a phone agent that needs a voice with a published realism result and one standard per-character rate, since Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365, and cloning is self-serve. Bland Speech costs $0.04 per 1,000 characters, pay as you go, with 50,000 free characters once for a new account.
Amazon Polly's Standard voices are $4 per million characters, a tenth of Bland Speech's $40, and a team converting a large archive of notifications where voice quality isn't the point has a fair reason to choose that tier. Bland publishes 199ms to first audio. An instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, and every clone is private to the customer's organization.
Bland Speech streams over an HTTP chunked endpoint and a WebSocket where a new context_id interrupts the turn that is playing, which is how a caller's interruption is handled. Output is PCM at 8 to 48 kHz or mu-law at 8 kHz, and performance tags such as [laughs] and [say warmly] shape a line. Pipecat and LiveKit Agents both have a Bland Speech integration, and Bland's text to speech API runs on two endpoints, POST /v2/tts and WSS /v2/tts/ws. Bland is SOC 2 Type II audited and HIPAA compliant with a signed BAA. Enterprise pricing is custom, set with Bland's sales team.
Pros
- Ranked 1st of 13 in the 4 August 2026 realism results
- Self-serve instant cloning from about 10 seconds
- Performance tags and barge-in on the WebSocket
- One standard rate, with no cloud project to configure
Cons
- More than a notification archive needs, because Bland Speech is built for live calls
The Bland Speech vs Google text to speech and Bland Speech vs Amazon Polly pages set each cloud against Bland, and the text to speech hub and rival vs rival family hold the rest. Bland Speech has the studio.
Frequently asked questions
Is Google text to speech better than Amazon Polly?
Google is better for expressive, tag-directed speech and Amazon Polly is better for a flat low price. Google's Gemini 2.5 Pro TTS ranked 5th of 13 in Design Arena's 4 August 2026 results, and Polly wasn't in those results.
Which is cheaper, Google text to speech or Amazon Polly?
At the oldest tier they cost the same, $4 per million characters for Google's WaveNet and Standard voices and for Polly's Standard voices. At the newer tier they match again, with Google's Chirp 3: HD and Polly's Generative both at $30 per million characters, and Google's first 1 million characters each month free against Polly's 100,000 a month for the first 12 months.
Can Google text to speech or Amazon Polly clone a voice?
Only in limited ways: Google's route is in Preview and Amazon's goes through an account manager. Google closed new access requests to Chirp 3 Instant Custom Voice and points to Gemini 3.8 voice replication, which is in Preview, and AWS builds a Brand Voice through a team's AWS Account Manager. Bland Speech's instant clone takes one clean sample of about 10 seconds.
Do Google text to speech and Amazon Polly support mu-law for phone calls?
Yes. Google offers MULAW and ALAW encodings, and Polly outputs mu-law and a-law at 8,000 Hz. Bland Speech offers mu-law at 8 kHz and PCM from 8 to 48 kHz.
Are Google text to speech and Amazon Polly HIPAA compliant?
Both are covered by a Business Associate Agreement. Google includes Text-to-Speech among the products under Google Cloud's BAA, and AWS calls Polly a HIPAA Eligible Service under the AWS BAA. Bland is HIPAA compliant with a signed BAA.
How fast are Google text to speech and Amazon Polly?
Google calls its streaming "ultra-low-latency speech" and AWS says Polly can "stream audio to your users in near real time." Bland publishes 199ms to first audio, and a team measures all three on its own text before choosing.
