GUIDE

Alexa skill custom voice: play your own audio

An Alexa skill speaks with Alexa’s voice or one of the Amazon Polly voices. For a quiz host, a story narrator, game characters or a brand voice, that is not enough, and SSML has an answer: the audio tag plays an MP3 file of your own in the middle of a response.

The free SuperClonica engine voices those files offline, in a stock voice, your own trained voice or a character. This guide shows Amazon’s exact file requirements, how to produce a compliant MP3 straight from Python and how to mix your audio with Alexa’s own speech.

Browse voices Install the SuperClonica engine

How custom audio works in Alexa

Let Alexa’s own voice read what changes, such as a score or a name, and play your files for everything that stays the same.

  • The SSML audio tag plays an MP3 from an HTTPS URL inside a response.
  • Amazon requires an MPEG version 2 MP3 at 48 kbps, with a sample rate of 16000, 22050 or 24000 Hz.
  • All audio in one response may last up to 240 seconds, 90 in a reprompt, with no more than five files.
  • SuperClonica speaks 31 languages; Chinese is not supported yet.

Hear the voices free online

On the page of any voice, type a line of your skill, 30 to 100 characters, choose the language and press Voice it. In a few seconds you hear whether it fits your host, narrator or character, with no account.

Make an Alexa-ready MP3

Keep each file short and focused: a greeting, a question, a sound cue. Short files load faster and are easier to reuse.

  • Install the engine: python -m pip install superclonica-cpu.
  • Voice a line and write it in Amazon’s format in one go: tts = Superclonica(); audio, rate = tts.synthesize("Welcome back! Your quiz starts now.", "F1", "en"); sf.write("welcome.mp3", resample_poly(audio, 24000, rate), 24000, format="MP3", bitrate_mode="CONSTANT", compression_level=0.75).
  • With soundfile 0.14 and libsndfile 1.2, which come with the engine, compression_level 0.75 gives a constant 48 kbps MPEG-2 file at 24 kHz.
  • Check every file before you upload it, with ffprobe or the file command.

Host it and play it

Host the files over HTTPS with a valid, trusted certificate, for example in an S3 bucket; self-signed certificates are rejected. Requests for the files are not authenticated, so they must not contain anything customer-specific or sensitive. In the response, put the audio tag with the file’s URL into the SSML string you pass to speak() in the ASK SDK, before or after Alexa’s own words.

Voices that pass certification

Amazon reviews every skill before certification. Use voices you have the rights to: the ten stock voices of the model, M1 to M5 and F1 to F5, your own voice trained in the SuperClonica studio from a recording of a speaker who agreed, or a custom voice ordered on the exchange. Fan-made character voices from the catalog are for personal projects, not for a published skill, and no voice should imitate a real person.