README
Sympatheia-18k
Sympatheia-18k is an emotion-aware spoken dialogue dataset for empathetic speech synthesis research. It contains 18,000 query–response pairs across 12 emotion categories, each accompanied by synthesized audio and text transcripts.
Dataset Structure
| Subset | Unique Queries | Responses | Description |
|---|---|---|---|
| Emotional | 8,400 train / 3,600 eval | 8,400 train / 3,600 eval | Emotional queries with emotionally-matched responses |
| Neutral | 350 train / 150 eval | 4,200 train / 1,800 eval | Neutral queries each paired with 12 emotion-targeted responses |
Files
{Emotional,Neutral}/audio/{train,eval}/: WAV audio files named{LABEL}_{INDEX}_{EMOTION}.wav{Emotional,Neutral}/metadata/text_pairs_{train,eval}.jsonl: query–response text pairs withquery_styleanduser_emotionlabels (see column descriptions below)encoded_{train,eval}.jsonl: audio tokens pre-encoded using the GLM-4-Voice decoder for use with the Sympatheia model, along with emotion labels and assigned valence–arousal values per emotion class{Emotional,Neutral}/metadata/emotion_va_mapping.json: per-class emotion → valence–arousal mapping (programmatically defined, not human-annotated)
Column Descriptions
Each text_pairs entry contains two emotion-related columns:
query_style: the TTS rendering style used to synthesize the query audio. In the Emotional subset this reflects the speaker's emotional delivery; in the Neutral subset this is alwaysNeutral.user_emotion: the underlying emotion attributed to the user, which determines the tone and content of the response. In the Emotional subset this is identical toquery_style. In the Neutral subset this is the inferred user emotion despite the neutral delivery (e.g. a neutrally-spoken message from someone who is actuallyFrustrated), and is the primary signal used to generate an appropriate empathetic response.
Related Project
This dataset was created for Sympatheia, an emotion-aware empathetic speech dialogue system.
Emotion Categories
12 discrete emotion classes: Angry, Anxious, Content, Disgusted, Excited, Frustrated, Happy, Neutral, Relaxed, Sad, Surprised, Tired
Contributors
6 commits
Similar repos
README
Sympatheia-18k
Sympatheia-18k is an emotion-aware spoken dialogue dataset for empathetic speech synthesis research. It contains 18,000 query–response pairs across 12 emotion categories, each accompanied by synthesized audio and text transcripts.
Dataset Structure
| Subset | Unique Queries | Responses | Description |
|---|---|---|---|
| Emotional | 8,400 train / 3,600 eval | 8,400 train / 3,600 eval | Emotional queries with emotionally-matched responses |
| Neutral | 350 train / 150 eval | 4,200 train / 1,800 eval | Neutral queries each paired with 12 emotion-targeted responses |
Files
{Emotional,Neutral}/audio/{train,eval}/: WAV audio files named{LABEL}_{INDEX}_{EMOTION}.wav{Emotional,Neutral}/metadata/text_pairs_{train,eval}.jsonl: query–response text pairs withquery_styleanduser_emotionlabels (see column descriptions below)encoded_{train,eval}.jsonl: audio tokens pre-encoded using the GLM-4-Voice decoder for use with the Sympatheia model, along with emotion labels and assigned valence–arousal values per emotion class{Emotional,Neutral}/metadata/emotion_va_mapping.json: per-class emotion → valence–arousal mapping (programmatically defined, not human-annotated)
Column Descriptions
Each text_pairs entry contains two emotion-related columns:
query_style: the TTS rendering style used to synthesize the query audio. In the Emotional subset this reflects the speaker's emotional delivery; in the Neutral subset this is alwaysNeutral.user_emotion: the underlying emotion attributed to the user, which determines the tone and content of the response. In the Emotional subset this is identical toquery_style. In the Neutral subset this is the inferred user emotion despite the neutral delivery (e.g. a neutrally-spoken message from someone who is actuallyFrustrated), and is the primary signal used to generate an appropriate empathetic response.
Related Project
This dataset was created for Sympatheia, an emotion-aware empathetic speech dialogue system.
Emotion Categories
12 discrete emotion classes: Angry, Anxious, Content, Disgusted, Excited, Frustrated, Happy, Neutral, Relaxed, Sad, Surprised, Tired
Similar repos
Linked in READMEs
Contributors
6 commits