atlithor/talromur3_without_emotions

Overview

This corpus is an emotion-less version of Talromur3_with_prompts. talromur_3_without_emotions is a prompt-labelled corpus that can be used for fine-tuning models, such as ParlerTTS.
The corpus consists of approximately 15,000 utterances, spoken by 7 named speakers in 6 different emotions (see more info here).

The dataset is an expanded version of Talromur-3: an Icelandic emotional speech corpus.
We have added natural-language descriptions of utterance-level pitch, speech monotony, speech quality, reverberation and speaking rate. Any information about
emotional content or intensity is omitted in this dataset.

Example description prompt:
Astrid's voice is clear and very close-sounding, with a monotone delivery. She speakes very slowly with a high-pitched tone.

We use DataSpeech to generate all features and the description prompts.
We replace g2p with an Espeak backend for phonemic transcription of the utterance texts in Icelandic.

Corpus information

Speakers

There are, currently, 7 speakers in Talromur-3. We create names for all speakers, which replace the speaker ID in the generated
description prompts. All speakers in Talromur-3 accounts for approximately the same amount of data.

Speaker ID	Speaker name	Speaker gender
F01	Astrid	female
F02	Freya	female
F03	Ingrid	female
F04	Frida	female
M01	Leif	male
M02	Anders	male
M03	Bjorn	male

Emotions

Utterances are spoken in 6 different emotions classes:

1. Neutral
2. Happy
3. Sad
4. Angry
5. Surprised
6. Helpful (child directed)

All non-neutral utterances are also assigned an emotional intensity label between 1 (very low) to 5 (very high). These features are, however, not represented
in the data as the description prompts for this corpus contain no information about the rendered emotion.

Example data

Below is an example row from the corpus annotated with explanations of features:

column	explanation	example data
speaker_id	Unique speaker ID, corresponds to original speaker ID in Talromur 3	F01
emotion	A textual description of the emotion of the utterance	medium intensity angry emotion (not used to generate the text_description)
text	The utterance text	Ég er líka hissa á að það skuli þurfa að flytja inn grænmeti.
language	language identfier	is
gender	the speaker's gender	female
utterance_pitch_mean	Mean F0	259.526337
utterance_pitch_std	Standard deviation of F0	52.141407
SNR	Signal-to-Noise Ratio	61.249115
c50	c50 speech clarity measurement	59.79673
phonemes	Phonemic transcription of the utterance target text	jɛːx ɛːr lˈiːɡa hˈɪssa aʊː aːð θaːð sɡˈyːlɪ θˈyrr#fa aːð flˈɪːdja ˈɪn ɡərˈaɪnmɛːdɪ
stoi	STOI objective intelligibility measurement	0.98991
si-sdr	Scale-invariant signal-to-distortion ratio measurement	23.737011
pesq	PESQ speech quality measurement	3.926996
speaking_rate	A textual description of the utterance speaking rate	very slowly
pitch	Textual description of the utterance-level pitch	high-pitch
noice	Textual description of the utterance-level noice	slightly clean
reverberation	Textual description of the utterance reverberation	very close-sounding
speech_monotony	Textual description of the utterance-level speech monotony	monotone
sdr_noise	A textual description of the utterance noice level	almost no noise
pesq_speech_quality	A textual description of the utterance speech quality	wonderful speech quality
text_description	An overall description prompt for the utterance.	Astrid delivers her monotone speech at a very slow pace, with a high-pitched voice. The recording is almost noiseless and her voice is very close-sounding.