{"url":"/method/fastspeech-2s","slug":"fastspeech-2s","name":"FastSpeech 2s","full_name":"FastSpeech 2s","full_name_withheld":false,"description_markdown":"**FastSpeech 2s** is a text-to-speech model that abandons mel-spectrograms as intermediate output completely and directly generates speech waveform from text during inference. In other words there is no cascaded mel-spectrogram generation (acoustic model) and waveform generation (vocoder). FastSpeech 2s generates waveform conditioning on intermediate hidden, which makes it more compact in inference by discarding the mel-spectrogram decoder.\r\n\r\nTwo main design changes are made to the waveform decoder. \r\n\r\nFirst, considering that the phase information is difficult to predict using a variance predictor, [adversarial training](https://paperswithcode.com/methods/category/adversarial-training)  is used in the waveform decoder to force it to implicitly recover the phase information by itself. \r\n\r\nSecondly, the mel-spectrogram decoder of [FastSpeech 2](https://paperswithcode.com/method/fastspeech-2) is leveraged, which is trained on the full text sequence to help on the text feature extraction. As shown in the Figure, the waveform decoder is based on the structure of [WaveNet](https://paperswithcode.com/method/wavenet) including non-causal convolutions and gated activation. The waveform decoder takes a sliced hidden sequence corresponding to a short audio clip as input and upsamples it with transposed 1D-convolution to match the length of audio clip. The discriminator in the adversarial training adopts the same structure in Parallel WaveGAN, which consists of ten layers of non-causal [dilated 1-D convolutions](https://paperswithcode.com/method/dilated-convolution) with [leaky ReLU](https://paperswithcode.com/method/leaky-relu) activation function. The waveform decoder is optimized by the multi-resolution STFT loss and the [LSGAN discriminator](https://paperswithcode.com/method/lsgan) loss following Parallel WaveGAN. \r\n\r\nIn inference, the mel-spectrogram decoder is discarded and only the waveform decoder is used to synthesize speech audio.","description_state":"present","introduced_year":null,"introduced_by":{"title":"FastSpeech 2: Fast and High-Quality End-to-End Text to Speech","paper":"/paper/fastspeech-2-fast-and-high-quality-end-to-end","first_author":"Yi Ren","n_authors":7,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/fastspeech-2-fast-and-high-quality-end-to-end"},"source":{"url":"https://arxiv.org/abs/2006.04558v8","title":"FastSpeech 2: Fast and High-Quality End-to-End Text to Speech","url_on_a_paper_host":true},"code_snippet_url":null,"code_snippet_url_on_a_code_host":false,"categories":[{"area":"Audio","area_id":"audio","collection":"Text-to-Speech Models","url":"/methods/category/text-to-speech-models","pwc_aliases":[]}],"n_papers_tagged":1,"archive_num_papers":1,"papers_newest_first":[{"paper":"/paper/fastspeech-2-fast-and-high-quality-end-to-end","title":"FastSpeech 2: Fast and High-Quality End-to-End Text to Speech","date":"2020-06-08","arxiv_id":"2006.04558","n_code_links":37,"syntology":{"ran":73,"of":119,"unverified":46,"pointer_only":33}}],"papers_shown":1,"tasks":[{"task":"/task/knowledge-distillation","name":"Knowledge Distillation","papers":1},{"task":"/task/speech-synthesis","name":"Speech Synthesis","papers":1},{"task":"/task/text-to-speech","name":"Text to Speech","papers":1},{"task":"/task/text-to-speech-synthesis","name":"Text-To-Speech Synthesis","papers":1},{"task":"/task/text-to-speech-1","name":"text-to-speech","papers":1}],"tasks_shown":5,"n_tasks":5,"usage_by_year":[{"year":"2020","papers":1}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/fastspeech-2s"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}