Amazon Polly's Standard voices cost $4 per million characters, a tenth of Bland Speech's $40, and AWS gives the first 5 million Standard characters free every month. A guide to leaving Polly that opened on savings would be wrong from its first line, because Bland Speech doesn't save money against Polly at any engine.
A team that does move has other reasons. A Polly Brand Voice goes through an AWS account manager, Polly's bidirectional streaming API is offered on one engine, and that Generative engine is one AWS says can sound slightly different after a model update. Bland keeps more text to speech migration guides beside its text to speech comparison hub.
Why would a team move from Amazon Polly to Bland Speech?
A team moves from Amazon Polly to Bland Speech when it wants a cloned voice it can create without an account-manager engagement, a WebSocket for streaming language model text, or a voice built for phone calls, and it accepts a higher per-character rate at every Polly engine, as of November 2026.
Cloning comes first. AWS says that for a Brand Voice, "please reach out to your AWS Account Manager or contact us for more information". Bland's instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio after Bland's consent and rights review, and every clone stays private to the customer's organization.
Streaming comes second. Polly's bidirectional streaming API, StartSpeechSynthesisStream, runs over HTTP/2 and is offered on the Generative engine in listed regions. Bland's WebSocket at /v2/tts/ws takes text as a language model writes it, and sending text under a new context_id cuts off the turn that was playing, which is how barge-in works. Bland publishes 199ms to first audio on a streamed request.
The third reason is stability across releases. AWS says that for Generative voices, "any updates to the training data and the model could result in slight variations to the way the voices sound." Bland's responses carry an x-model header naming the model a voice resolved to, and BTTS_V2 voices still synthesize alongside BTTS_V3, so a team can log which model produced each call.
Amazon Polly wasn't in Design Arena's Audio Realism Bench results of 4 August 2026, where Bland Speech v3 ranked first of 13 AI voices at 1365 and the human recording scored 1500, so no score from that benchmark places Polly's voices next to Bland's. A team that wants to know how they compare listens to its own lines in Speech Studio next to Polly's output.
Staying is a fair answer in some cases. A team whose prompts are short, fixed, and generated once can keep them on Polly Standard at $4 per million characters, and AWS says "you can cache and replay Amazon Polly's generated speech at no additional cost."
What maps to what when moving from Amazon Polly to Bland Speech?
Polly's four engines map to one Bland model family where the voice decides the model, Polly's StartSpeechSynthesisStream maps to Bland's WebSocket, and Polly's mu-law at 8,000 Hz maps to Bland's mulaw at 8,000 Hz.
What these rows mean
- Engine: Bland says a team doesn't choose a model, because the voice it passes decides the model, and the model list names
btts-3as current andbtts-2as older. A team picking a voice is picking a model. - Request size: AWS counts billed characters separately from total characters, which is why one limit has two numbers.
- Delivery: Newscaster style isn't available on Polly's Generative engine, so a team that uses it on other engines re-voices those lines with tags.
- Price: each side's published rate. Polly's free tier is 5 million Standard characters a month, and 1 million Neural, 500 thousand Long-Form, and 100 thousand Generative a month for the first 12 months.
The delivery rewrite is the longest part of a move for teams that lean on SSML. An SSML break becomes a pause marker written <|0.8|>, and each marker adds about a second of breath on top of its length. Whispering becomes the tag [say in a whisper], and Bland publishes the full trained set and says invented variants are "ignored or garbled", so each tag is copied exactly from that list. Bland's tags work on its current-generation voices only, which is the BTTS_V3 family.
Frameworks are the easy case. The Pipecat team maintains services for both AWS Polly and Bland, and Bland's installs with uv add "pipecat-ai[bland]". LiveKit Agents has Python plugins for both Amazon Polly and Bland, and Bland's installs with uv add "livekit-agents[bland]~=1.8".
What are the steps to switch from Amazon Polly to Bland Speech?
A switch takes six steps: build a voice table, rewrite SSML as plain text, change the audio request, pick an endpoint per use, regenerate cached prompts, and run both providers in parallel before cutover.
- A team builds a voice table with each Polly voice beside a Bland voice UUID.
GET /v1/voiceslists every voice a team can use, three built-in voices (Karen, Matthew, and River) are open to every organization, and Speech Studio renders a line with the same models the API uses. - A team rewrites each SSML document as plain text with Bland's tags and pause markers, and sets
expressivenessandstabilityper request. - A team changes the audio request. Polly's mu-law at 8,000 Hz becomes
"audio": { "encoding": "mulaw", "sample_rate": 8000 }, and Polly's 16,000 Hz PCM becomes"encoding": "pcm_s16le", "sample_rate": 16000. A team with no phone leg asks for 48,000 Hz, whichBTTS_V3renders natively. - A team picks an endpoint per use. Bland's own migration card for Amazon Polly describes the job as "Move synchronous synthesis or add token-level realtime streaming", which maps to
POST /v2/ttsfor complete text and the WebSocket for language model output. - A team regenerates its cached prompt library. Bland advises raising
stabilitywhen audio is cached or copy is A/B tested, because a higher value repeats itself closely. - A team routes a small share of calls to Bland, compares recordings of the same lines, and moves the rest once the results hold.
curl -X POST "https://api.bland.ai/v2/tts" \
-H "Authorization: Bearer $BLAND_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "Your appointment is confirmed for Tuesday at 3 PM.",
"voice": "29158307-9893-4149-8a75-bc9ce313d64e",
"audio": { "encoding": "mulaw", "sample_rate": 8000 }
}' \
--output prompt.ulaw
That request returns raw mu-law audio ready for a phone leg. The response headers report x-cost, x-model, and x-latency for the call, and Bland bills a $0.001 minimum on each request, so a very short prompt costs the minimum.
What does the move cost?
At 20 million characters a month, Polly Standard is $60, Neural is $320, Generative is $600, and Long-Form is $2,000, while Bland Speech is $800, and the $200 gap to Generative is the smallest one.
Those Polly figures use AWS's published per-million rates, with the 5 million free Standard characters taken off the Standard line and no free tier taken off the others. The Bland figure is $0.04 per 1,000 characters, pay as you go, with no monthly plan, and a new account gets 50,000 free characters once. Twenty million characters is roughly 370 hours of speech at Bland's own example of $0.36 for a 10-minute narration, which is 9,000 characters.
Standard is where the gap is widest, at about 13 times the cost with the free allowance counted. Polly's generative voices are the closer comparison for a live agent, and there Bland costs a third more.
What can't come along from Amazon Polly?
A Polly Brand Voice doesn't come along, because Bland voices are made on Bland from audio, and a team that built one through AWS clones again from the original recordings with the speaker's permission.
Bland says voice IDs are provider-specific and can't be reused, so a voice table is built once and kept. Compliance gets compared, not copied. AWS says Polly is a HIPAA Eligible Service covered under the AWS Business Associate Addendum, and that Polly "is certified for use with regulated workloads for HIPAA" and PCI DSS. Bland is SOC 2 Type II audited, HIPAA compliant with a signed BAA that it offers on Enterprise, and self-attested to PCI DSS v4.0.1. Bland is also FedRAMP 20x Class C certified at Moderate, granted 17 November 2026, and says speech synthesis runs inside the FedRAMP boundary. Enterprise pricing is custom, set with Bland's sales team.
The Bland Speech vs Amazon Polly head-to-head, the Amazon Polly alternatives list, and the Amazon Polly review cover the wider choice. The text to speech API page lists each endpoint, and Bland Speech has the studio.
Frequently asked questions
Is Bland Speech cheaper than Amazon Polly?
No. Bland Speech is $0.04 per 1,000 characters ($40 per million), and Amazon Polly is $4 for Standard, $16 for Neural, $30 for Generative, and $100 for Long-Form per million characters. AWS also gives 5 million Standard characters free every month, and Bland gives a new account 50,000 free characters once.
Which Polly engine does Bland Speech replace?
All four. Bland has one model family, and the voice a team picks decides the model, so there is no engine setting to match. Current voices run on BTTS_V3, which renders natively at 48 kHz, and older BTTS_V2 voices still synthesize.
Can a Polly Brand Voice move to Bland Speech?
No, a Polly Brand Voice can't move as a voice. A team clones again from its original recordings, with an instant clone needing one clean sample of about 10 seconds and a professional clone needing 30 minutes or more plus Bland's consent and rights review.
Does Polly's 8 kHz mu-law output carry over?
Yes. Polly's mu-law at 8,000 Hz maps to Bland's mulaw at 8,000 Hz, and Bland also returns pcm_s16le at 8,000 to 48,000 Hz for a leg that takes linear PCM.
How do Polly SSML pauses translate to Bland Speech?
A Polly break becomes a Bland pause marker, written <|N|> with N from 0.1 to 10.0 seconds. A marker above 10.0 fails the whole request instead of being clamped, and each marker adds about a second of breath on top of its length.
Can Pipecat or LiveKit apps switch to Bland Speech?
Yes. Pipecat ships a Bland service maintained by its own team, and LiveKit Agents has a Bland plugin for Python, so a voice agent on either framework swaps its text to speech service without leaving the framework.
Does Polly's HIPAA and PCI coverage carry over?
Not automatically. AWS covers Polly under its BAA and says Polly is certified for HIPAA and PCI DSS workloads, and Bland is HIPAA compliant with a signed BAA and self-attested to PCI DSS v4.0.1. A team with payment audio compares those statuses against its own assessor's requirements before moving.
