FastSpeech 2
A non-autoregressive TTS model that uses explicit variance modeling to separately predict duration, pitch, and energy, enabling fast and controllable speech synthesis.
Timeline (last 90 days)
A modern TTS pipeline is typically three-stage — "text front end + acoustic model (such as FastSpeech 2) + vocoder (such as HiFi-GAN)" — and voice cloning adds speaker embeddings or reference audio prompts on top of that.
After 2020, end-to-end TTS models based on the Transformer architecture (such as FastSpeech 2 and VITS) integrated text analysis, prosody prediction, and waveform generation into a single model.
All Facts (2)
A modern TTS pipeline is typically three-stage — "text front end + acoustic model (such as FastSpeech 2) + vocoder (such as HiFi-GAN)" — and voice cloning adds speaker embeddings or reference audio prompts on top of that.
50%UnverifiedAfter 2020, end-to-end TTS models based on the Transformer architecture (such as FastSpeech 2 and VITS) integrated text analysis, prosody prediction, and waveform generation into a single model.
50%