--- license: cc-by-4.0 task_categories: - text-to-speech language: - en - te tags: - tts - speech - indian-languages - telugu - indian-english - emotion - single-speaker pretty_name: Indian English + Telugu Single-Speaker TTS (emotion-tagged) size_categories: - n<1K configs: - config_name: indian_english data_files: - split: train path: indian_english/train-* - split: validation path: indian_english/validation-* - split: test path: indian_english/test-* - config_name: telugu data_files: - split: train path: telugu/train-* - split: validation path: telugu/validation-* - split: test path: telugu/test-* dataset_info: - config_name: indian_english features: - name: audio dtype: audio: sampling_rate: 24000 - name: text dtype: string - name: normalized_text dtype: string - name: language dtype: string - name: language_code dtype: string - name: emotion dtype: string - name: style dtype: string - name: emotion_confidence dtype: float32 - name: tag_source dtype: string - name: speaker_id dtype: string - name: gender dtype: string - name: accent dtype: string - name: duration dtype: float32 - name: snr_db dtype: float32 - name: dnsmos_ovrl dtype: float32 - name: dnsmos_sig dtype: float32 - name: dnsmos_bak dtype: float32 - name: dnsmos_pass dtype: bool - name: squim_stoi dtype: float32 - name: squim_pesq dtype: float32 - name: squim_sisdr dtype: float32 - name: mms_align_score dtype: float32 - name: overlap_flag dtype: bool - name: ser_emotion dtype: string - name: valence dtype: float32 - name: arousal dtype: float32 - name: dominance dtype: float32 - name: source_video_id dtype: string - name: source_url dtype: string - name: source_channel dtype: string - name: license dtype: string - name: segment_start dtype: float32 - name: segment_end dtype: float32 - name: sample_rate dtype: int32 splits: - name: train num_bytes: 78049373.0 num_examples: 151 - name: validation num_bytes: 4637837.0 num_examples: 9 - name: test num_bytes: 3842320.0 num_examples: 9 download_size: 84985707 dataset_size: 86529530.0 - config_name: telugu features: - name: audio dtype: audio: sampling_rate: 24000 - name: text dtype: string - name: normalized_text dtype: string - name: language dtype: string - name: language_code dtype: string - name: emotion dtype: string - name: style dtype: string - name: emotion_confidence dtype: float32 - name: tag_source dtype: string - name: speaker_id dtype: string - name: gender dtype: string - name: accent dtype: string - name: duration dtype: float32 - name: snr_db dtype: float32 - name: dnsmos_ovrl dtype: float32 - name: dnsmos_sig dtype: float32 - name: dnsmos_bak dtype: float32 - name: dnsmos_pass dtype: bool - name: squim_stoi dtype: float32 - name: squim_pesq dtype: float32 - name: squim_sisdr dtype: float32 - name: mms_align_score dtype: float32 - name: overlap_flag dtype: bool - name: ser_emotion dtype: string - name: valence dtype: float32 - name: arousal dtype: float32 - name: dominance dtype: float32 - name: source_video_id dtype: string - name: source_url dtype: string - name: source_channel dtype: string - name: license dtype: string - name: segment_start dtype: float32 - name: segment_end dtype: float32 - name: sample_rate dtype: int32 splits: - name: train num_bytes: 77429294.0 num_examples: 140 - name: validation num_bytes: 4865171.0 num_examples: 8 - name: test num_bytes: 4432899.0 num_examples: 8 download_size: 85838444 dataset_size: 86727364.0 --- # Indian English + Telugu Single-Speaker TTS Dataset (emotion-tagged) Clean audio clips sourced from YouTube, transcribed with **Sarvam** ASR, segmented with diarization, and labeled with emotion/style tags. Built as a data-quality / curation exercise. > **"Single-speaker"** means **each clip contains exactly one speaker** (verified by > diarization and speaker-embedding similarity). The dataset spans **11 distinct speakers > total** (5 English, 6 Telugu), tracked via `speaker_id`. ## Contents - **Indian English** (`indian_english`): 30.05 min, 142 clips, 5 speakers; emotions: {'neutral': 31, 'calm': 31, 'sad': 30, 'excited': 30, 'angry': 12, 'fearful': 5, 'happy': 3} - **Telugu** (`telugu`): 30.25 min, 140 clips, 6 speakers; emotions: {'calm': 25, 'neutral': 24, 'angry': 24, 'excited': 24, 'sad': 24, 'fearful': 9, 'happy': 7, 'surprised': 3} Total: **60.3 minutes**. ## Evaluation (evidence, not just claims) - **Single-speaker check** (ECAPA-TDNN embeddings): same-speaker cosine 0.74 vs different-speaker 0.21 (separation 0.52, verification AUC 0.96 / EER 9.1%; 0/11 speakers flagged). - **Transcript reliability**: English cross-ASR agreement with Whisper = 6.8% WER / 4.5% CER (n=40) — strong. Realtime ASR language-ID matched the target language on 100% of EN and 100% of TE clips. Telugu cross-ASR is not a valid proxy (Whisper is weak in Telugu); Telugu transcripts are best audited by human review. - **Emotion-tag reliability** (sarvam-30b vs sarvam-105b on 120 clips): 65% agreement, Cohen's κ 0.55. - **Phoneme coverage**: English 39 (100%), Telugu 45 (90%). See the project report (GitHub repo) for full methodology and figures. ## Schema `audio` (24 kHz mono), `text`, `normalized_text`, `language`, `language_code`, `emotion` (neutral, happy, sad, angry, excited, calm, fearful, surprised), `style` (narrative, conversational, formal, expressive, whisper), `emotion_confidence`, `tag_source` (`auto`/`human`), `speaker_id`, `duration`, `snr_db`, `source_video_id`, `source_url`, `source_channel`, `license`, `segment_start`, `segment_end`, `sample_rate`. ## How it was built 1. Curated single-speaker YouTube sources (audiobooks, lectures, news, storytelling). 2. **Sarvam batch STT** (`saaras:v3`) with diarization + timestamps for structure. 3. Silence-snapped segmentation into 3–25 s clips (single speaker only). 4. **Sarvam realtime STT** (`saarika:v2.5`) per clip for clip-accurate transcripts. 5. Automated quality gates (clipping, SNR, silence, music/noise bed, ASR confidence, dedup). 6. Hybrid emotion tagging: per-speaker-normalized acoustic features + Sarvam LLM, with an acoustic whisper override. 7. Human review (listen, fix transcripts, relabel) — **human labels override automated ones**. 8. Light loudness normalization (dynamics preserved), balanced emotion selection. ## Audio 24 kHz mono WAV. Loudness lightly normalized (~-20 LUFS, peak −1 dBFS) WITHOUT limiting, so the prosodic dynamics that carry emotion are preserved. ## Ethics & licensing Sourced from YouTube for research; clips are short and transformative. Per-clip provenance (`source_url`, `source_channel`, `license`) is retained. Respect the original creators' rights; remove clips on request. ## Limitations Emotion tags are heuristic (acoustic + LLM, partly human-verified) and may be imperfect for subtle prosody. See the project report for iteration notes.