README.md

Call Transcript Intent Dataset

Multimodal Hindi/Hinglish customer utterance dataset for loan/EMI/payment call intent classification.

Dataset Summary

Metric	Value
Total examples	139,348
Total audio duration	51.04 h
Number of intents	17

Split Statistics

Split	Examples	Duration	Hours
train	126,848	2755.14 min	45.92 h
validation	10,000	219.03 min	3.65 h
eval	2,500	88.37 min	1.47 h

Class Distribution

Intent	Total	Train	Validation	Eval
UNCLEAR_INPUT	75,051	68,669	5,524	858
AFFIRMATIVE_ACKNOWLEDGEMENT	37,719	34,652	2,710	357
CONTINUE_CONVERSATION	6,366	5,768	477	121
PAY_LATER_AGREE	3,971	3,621	220	130
NEGATIVE_ACKNOWLEDGEMENT	3,641	3,260	254	127
NO_PAYMENT_REASON	2,963	2,680	210	73
BACKCHANNEL_OR_NOISE	2,096	1,907	167	22
PAID_ALREADY	1,836	1,605	114	117
THIRD_PARTY_UNAVAILABLE	1,035	878	63	94
IDENTITY_CONFIRMED	1,027	846	75	106
PAY_NOW_AGREE	819	663	48	108
END_CALL	744	606	33	105
CALL_DEFER	622	579	31	12
REFUSE_TO_PAY	462	324	29	109
THIRD_PARTY_AVAILABLE	367	322	29	16
WRONG_NUMBER	343	286	15	42
DO_NOT_CALL	286	182	1	103

Columns

• audio
• id
• oid
• conversation_id
• recording_url
• chunk_index
• start_ms
• end_ms
• duration_s
• transcript
• transcript_reviewed
• intent