dataset_info:
config_name: speech
features:
- name: input_ids
sequence: int32
splits:
- name: read
num_examples: 11563
- name: improv_2048
num_examples: 4010
- name: improv_4096
num_examples: 2311
- name: improv_6144
num_examples: 1595
- name: improv_8192
num_examples: 1218
configs:
- config_name: speech
data_files:
- split: read
path: read.parquet
- split: improv_2048
path: improv_2048.parquet
- split: improv_4096
path: improv_4096.parquet
- split: improv_6144
path: improv_6144.parquet
- split: improv_8192
path: improv_8192.parquet
license: cc-by-nc-4.0
language:
- en
pretty_name: The Expresso Dataset
Info about this dataset
This is the complete (sans irregular styles) expresso dataset, merged from the following sources:
- improvised conversational: nytopop/expresso-improv
- read speech: ylacombe/expresso
All audio was resampled to 24khz and encoded with SNAC as orpheus codebooks. Conversational turns from the improvised subset are arranged as multi-turn windows.
Each turn is tokenized in orpheus layout. Plaintext prefixes are used to indicate speaker / style information of each example. That is, text prompts of each turn are of the form:
speaker: [style] transcript
Except for the default, essentials, and emphasis styles, for which the [style] tag is omitted (all are variations of default in the same delivered style with in-text markup).
Speaker IDs are renamed as follows:
| speaker_id | speaker |
|---|---|
| ex01 | jack |
| ex02 | lisa |
| ex03 | bert |
| ex04 | emma |
The read split contains read speech examples (10.73 hours, 11,563 turns, 3,485,769 tokens).
The improv_2048 split contains improvised conversational examples <= 2048 tokens (20.23 hours, 14,818 turns, 6,523,432 tokens).
The improv_4096 split contains improvised conversational examples <= 4096 tokens (24.23 hours, 15,276 turns, 7,762,324 tokens).
The improv_6144 split contains improvised conversational examples <= 6144 tokens (25.20 hours, 15,339 turns, 8,064,934 tokens).
The improv_8192 split contains improvised conversational examples <= 8192 tokens (25.48 hours, 15,351 turns, 8,151,443 tokens).
Style stats by split
| Style | Read (turns) | Improv (turns) | Total (turns) |
|---|---|---|---|
| confused | 1515 | 567 | 2082 |
| default | 2475 | 1213 | 3688 |
| enunciated | 1500 | 762 | 2262 |
| happy | 1519 | 639 | 2158 |
| laughing | 1519 | 804 | 2323 |
| sad | 1518 | 1082 | 2600 |
| whisper | 1517 | 806 | 2323 |
| disgusted | - | 1073 | 1073 |
| sympathetic | - | 1067 | 1067 |
| narration | - | 394 | 394 |
| projected | - | 1195 | 1195 |
| fast | - | 1065 | 1065 |
| angry | - | 695 | 695 |
| sarcastic | - | 757 | 757 |
| calm | - | 880 | 880 |
| desire | - | 195 | 195 |
| nonverbal | - | 327 | 327 |
| fearful | - | 394 | 394 |
| awe | - | 648 | 648 |
| bored | - | 274 | 274 |
| sleepy | - | 514 | 514 |
The Expresso Dataset
[paper] [demo samples] [Original repository]
Introduction
The Expresso dataset is a high-quality (48kHz) expressive speech dataset that includes both expressively rendered read speech (8 styles, in mono wav format) and improvised dialogues (26 styles, in stereo wav format). The dataset includes 4 speakers (2 males, 2 females), and totals 40 hours (11h read, 30h improvised). The transcriptions of the read speech are also provided.
You can listen to samples from the Expresso Dataset at this website.
Audio Quality
The audio was recorded in a professional recording studio with minimal background noise at 48kHz/24bit. The files for read speech and singing are in a mono wav format; and for the dialog section in stereo (one channel per actor), where the original flow of turn-taking is preserved.
License
The Expresso dataset is distributed under the CC BY-NC 4.0 license.
Reference
For more information, see the paper: EXPRESSO: A Benchmark and Analysis of Discrete Expressive Speech Resynthesis, Tu Anh Nguyen*, Wei-Ning Hsu*, Antony D'Avirro*, Bowen Shi*, Itai Gat, Maryam Fazel-Zarani, Tal Remez, Jade Copet, Gabriel Synnaeve, Michael Hassid, Felix Kreuk, Yossi Adi⁺, Emmanuel Dupoux⁺, INTERSPEECH 2023.