A new generative engine and three voices are now generally available on Amazon Polly | Amazon Web Services

Today, we are announcing the general availability of the generative engine of Amazon Polly with three voices: Ruth and Matthew in American English and Amy in British English. The new generative engine was trained with publicly available and proprietary data and a variety of voices, languages, and styles. It performs with the highest precision to render context-dependent prosody, pausing, spelling, dialectal properties, foreign word pronunciation, and more.

Amazon Polly is a machine learning (ML) service that converts text to lifelike speech, called text-to-speech (TTS) technology. Now, Amazon Polly includes high-quality, natural-sounding humanlike voices in dozens of languages, so you can select the ideal voice and distribute your speech-enabled applications in many locales or countries.

With Amazon Polly, you can select various voice options, including neural, long-form, and generative voices, which deliver ground-breaking improvements in speech quality and produce human-like, highly expressive, and emotionally adept voices. You can store speech output in standard formats like MP3 or OGG, adjust the speech rate, pitch, or volume with Speech Synthesis Markup Language (SSML) tags, and quickly deliver lifelike voices and conversational user experiences with consistently fast response times.

What’s the new generative engine?
Amazon Polly now supports four voice engines: standard, neural, long-form, and generative voices.

Standard TTS voices, introduced in 2016 use traditional concatenative synthesis. This method strings together the phonemes of recorded speech, producing very natural-sounding synthesized speech. However, the inevitable variations in speech and the techniques used to segment the waveforms limit the quality of speech.

Neural TTS (NTTS) voices, introduced in 2019, use a sequence-to-sequence neural network that converts a sequence of phonemes into spectrograms and a neural vocoder that converts the spectrograms into a continuous audio signal. The NTTS produces even higher quality humanlike voices than its standard voices.

Long-form voices, introduced in 2023, are developed with cutting-edge deep learning TTS technology and designed to captivate listeners’ attention for longer content, such as news articles, training materials, or marketing videos.

In February 2024, Amazon scientists introduced a new research TTS model called Big Adaptive Streamable TTS with Emergent abilities (BASE). With this technology, the Amazon Polly generative engine is able to create humanlike synthetically generated voices. You can use these voices as a knowledgeable customer assistant, a virtual trainer, or an experienced marketer.

Here are the new generative voices:

$ aws polly describe-voices --output json --region us-east-1 
| jq -r '.Voices[] | select(.SupportedEngines | index("generative")) | .Name'

Matthew
Amy
Ruth

$ aws polly synthesize-speech --output-format mp3 --region us-east-1 
  --text "Hello. This is my first generative voices!" 
  --voice-id Matthew --engine generative hello.mp3