TTSMP3

TTSMP3 runs two converters with separate daily allowances. The Regular Voices page, labelled Powered by AWS Polly, accepts about 3,000 characters or 375 words a day, which is also the most you can paste into one conversion. The AI Voices page, labelled Powered by OpenAI Text-To-Speech, stops at about 1,000 characters or 125 words a day and shows a counter of what you have used.

Because the site counts characters rather than words, markup counts too. A dialogue with pause and emphasis tags uses part of the allowance on tags that are never spoken, so strip unneeded tags from long scripts before converting.

The three paid options are sold in euros. The €5.00 pass gives 1,000,000 characters that must be used within 24 hours and does not renew. The monthly plan at €9.17 gives 1,000,000 characters each month, and the yearly plan at €90.83 gives 10,000,000 characters for the year and is the only one that includes the API.

Editorial assessment

A 60,000-character course script would take 20 days of free Polly allowance, but a single 24-hour pass covers it if the script is final before you buy. Choose the monthly plan when new narration arrives every week, and the yearly pass only if you need the API or expect to use most of the 10 million characters; per million characters it costs almost the same as paying monthly.

The two voice sets are controlled differently. Regular Voices read 28 languages and accents with a fixed speaker per language, and take SSML tags for pauses, emphasis, speaking rate, pitch and whispering. AI Voices offer nine voices such as Alloy, Nova and Onyx, detect the language on their own from a list of 57, and take a written instruction describing identity, affect, tone, emotion and pronunciation.

AI Voices cannot change pitch or tone on demand; the model infers inflection from punctuation and context. The FAQ says neither voice set can be downloaded for offline use, and new languages or accents depend on what Amazon and OpenAI release rather than on TTSMP3.

Regular Voices accept Amazon Polly SSML, so you can insert breaks, emphasis, rate and pitch changes, and a whispered effect directly in the text box. To switch speakers inside one conversion, start each line with a tag such as [speaker:Brian] or [speaker:Emma], which renders a short conversation in one MP3.

Two details decide whether a conversion succeeds:

• Speaker names in dialogue tags must be typed without accent marks, as the converter page instructs for the French and Spanish voice names.
• Every SSML tag must be closed, and stray < or > characters in ordinary text are read as broken tags and make the conversion fail.

API access comes only with the yearly pass. After buying, you email TTSMP3 for a key; the documentation then describes a GET or POST request to api.ttsmp3.com with the key, a speaker name and the text. A successful response includes the characters used, the remaining quota and a URL to the MP3 file on the TTSMP3 server.

Repeating a text that was already synthesized returns the cached file without deducting characters, and a request that would exceed the quota is refused with status 320 instead of converting part of the text. The documented speaker list is the Regular Voices set; the documentation does not describe AI Voices in the API.

The terms name EPSILON WEBCRAFTS LLC as operator and Kentucky, USA as governing law, but they show an update line without a month and say nothing about refunds, voice rights or how long generated files are kept. The purchase FAQ covers cancellation, but refunds are not addressed. The free converter shows ads and asks for consent to share device data with 210 advertising partners; paid access removes ads.

The FAQ is clearer on rights than the terms: MP3s from both free and paid use may go into commercial and non-commercial projects, including YouTube videos, apps and e-learning, without crediting TTSMP3. Keep a copy of the FAQ page with your project records, because the terms themselves do not repeat that licence.

1. Write the dialogue with [speaker:Brian] and [speaker:Amy] tags and close every SSML tag.
2. Convert it once, note the date, the voice pair and the characters counted, and download the MP3.
3. Listen for the speaker switches, both breaks, the emphasis and the pronunciation of the names, date and price.
4. Repeat the same text with one AI Voice using a short tone instruction and compare the pronunciation of the same items.

Accept the result if every speaker switch and break lands in place and the names, date and price are read correctly. Reject it if a tag is spoken aloud, the conversion fails on markup or a name is mispronounced in both voice sets. No audio result is reported until the conversion is actually run.

If a free weekly quota suits your volume better than a daily one, compare TTSMaker; for longer studio-style voiceovers synced to video, see Murf.