Can You Use Amazon Polly for Audiobooks? (2026 Honest Guide)

Amazon Polly is one of the cheapest ways to get neural-quality speech — roughly $4 to $16 per million characters. For a full book that is just a few dollars of synthesis. So why doesn't everyone use it for audiobooks? Because Polly is an API, not a product. It hands you raw audio and nothing else.

This guide walks through what it actually takes to make an audiobook with Polly, where it breaks down, and when a purpose-built tool saves you days of work.

What Amazon Polly Is Good At

• Cost — Neural voices are extremely cheap per character. A 60,000-word book (~350k characters) costs a few dollars to synthesize.
• Reliability — It is AWS infrastructure. It does not go down mid-project.
• Language coverage — Dozens of neural voices across many languages.
• SSML control — You can fine-tune pronunciation, pauses, and emphasis with markup.

The Real Polly Audiobook Workflow

Here is what “just use Polly” actually involves:

1. Set up AWS. Create an account, configure IAM credentials, install the SDK or CLI.
2. Chunk the text. Polly's synchronous API caps at 3,000 billable characters per request; long text needs the asynchronous task API and S3. Either way, you script the chunking.
3. Synthesize in batches. Loop through chunks, handle rate limits and retries, save each audio file.
4. Stitch and split. Concatenate chunks back into chapters, insert pauses, and split the book into per-chapter files.
5. Master for ACX. Convert to 44.1kHz/192kbps MP3, normalize loudness, add room tone, check peak levels. None of this is built in.

None of these steps is hard for an engineer. All of them together turn a “cheap” audiobook into a weekend project — and a fragile one to repeat for the next book.

Where Polly Falls Short for Books

• No interface — Everything is code. Non-developers are stuck.
• No chapter awareness — Polly sees a stream of text, not a structured book.
• No ACX output — You handle every Audible technical requirement manually.
• Voice fatigue — Neural voices are good in short bursts; over many hours, the best long-form engines hold up better.
• No voice cloning for general users — narrating in your own voice is not on the table.

The No-Code Alternative

AudioBookByMe gives you long-form-tuned voices (via the open-source Kokoro engine) without any of the AWS plumbing. You upload PDF, EPUB, Word, or scanned pages; it detects chapters and exports ACX-compliant MP3s. The free tier lets you start at no cost, and voice cloning is included on the Pro plan ($19.99/month).

Further Reading

• Can You Use Google Text-to-Speech for Audiobooks? (Honest Answer)
• Kokoro TTS: The Free Open-Source Voice Engine Explained
• Text-to-Speech for Audiobooks: AI Voices vs Human Narrators