Marathi Conversational Speech

Natural, open-domain Marathi conversations recorded between consenting adult speakers, with each side of the call captured on its own channel for clean two-party diarization. Sessions are unscripted, so models hear real turn-taking, overlap, backchannels, and spontaneous speech rather than read prompts. Speakers span Maharashtra across urban and rural settings.

Every session is collected from paid contributors with signed consent, transcribed and checked by native speakers, and screened for personal information before delivery.

Highlights

• Stereo capture with left and right speaker separation for clean two-party diarization
• Natural, open-domain Marathi dialogue between consenting adults, not read scripts
• Verbatim native-speaker transcripts with speaker labels and utterance timestamps
• Utterance-level emotion and sentiment annotations available on request
• Consent-cleared and PII-reviewed before delivery

Topic and speaker coverage

Balanced across speakers, regions, and everyday topics. Speakers span Maharashtra across urban and rural settings. Coverage extends to specific domains, accents, or demographic targets on request.

Capture and format

Stereo WAV at 16 kHz or higher (48 kHz available), one speaker per channel, recorded in low-noise environments. Both sides are time-aligned and segmented into utterances, with the full session retained rather than trimmed highlights.

Annotations

Verbatim native-speaker transcripts with speaker IDs and utterance timestamps as standard, plus optional emotion, sentiment, topic, and code-switching labels for Marathi on request.

Provenance

• Paid contributors with signed consent
• Native-speaker transcription and quality review
• PII reviewed and redacted before delivery
• Per-recording audit trail and licensable usage rights

Use cases

• ASR and speaker-diarization training for Marathi
• Conversational TTS and voice-agent development
• Emotion and sentiment recognition from speech
• Spontaneous-speech and code-switching research