id stringlengths 16 16 | language stringclasses 2
values | audio_array listlengths 110k 584k | sampling_rate int32 22.1k 22.1k | transcript stringlengths 47 550 | duration_sec float32 5 26.5 | emotion stringclasses 9
values | style stringclasses 8
values | manually_verified bool 1
class | needs_manual_review bool 2
classes | confidence float32 0.38 0.95 |
|---|---|---|---|---|---|---|---|---|---|---|
en_src001_seg001 | en-IN | [0.00274658203125,0.00244140625,0.0013427734375,0.0006103515625,0.00103759765625,0.00079345703125,0.(...TRUNCATED) | 22,050 | "Hello everyone, I hope the energies are still lifted and we had a great panel. I think there was a (...TRUNCATED) | 8.9 | neutral | conversational | false | false | 0.75 |
en_src001_seg002 | en-IN | [-0.00140380859375,-0.00164794921875,-0.00067138671875,-0.0025634765625,-0.0054931640625,-0.00628662(...TRUNCATED) | 22,050 | "Uh, before that, we move to our live pitching. We have Sourav again on stage. He wants to share som(...TRUNCATED) | 7.4 | neutral | conversational | false | false | 0.75 |
en_src001_seg004 | en-IN | [0.00079345703125,0.000732421875,0.000579833984375,0.000762939453125,0.00091552734375,0.000946044921(...TRUNCATED) | 22,050 | Hello everyone! Itne saare logon mein itni chhoti si hello. | 5.3 | excited | formal | false | true | 0.65 |
en_src001_seg006 | en-IN | [0.00030517578125,0.00030517578125,0.000152587890625,0.00006103515625,0.000152587890625,0.0007629394(...TRUNCATED) | 22,050 | "Very nice. Okay. So, as Dr. Munish Jindal said, we need to meet Connect today, okay, now. There can(...TRUNCATED) | 8.9 | excited | formal | false | true | 0.65 |
en_src001_seg007 | en-IN | [-0.00079345703125,-0.000732421875,-0.00048828125,-0.00030517578125,-0.000274658203125,-0.0003051757(...TRUNCATED) | 22,050 | "So we have created something for you. Open your WhatsApps. You might have got a message from someth(...TRUNCATED) | 7.4 | neutral | informational | false | false | 0.75 |
en_src001_seg012 | en-IN | [-0.010986328125,-0.01068115234375,-0.0101318359375,-0.009246826171875,-0.00823974609375,-0.00723266(...TRUNCATED) | 22,050 | "Make sure you have enough keywords in your bio, that you are a founder, investor, whatever you are (...TRUNCATED) | 6.899955 | excited | formal | false | true | 0.65 |
en_src001_seg013 | en-IN | [-0.0003662109375,-0.000244140625,-0.0003662109375,-0.0006103515625,-0.000762939453125,-0.0005798339(...TRUNCATED) | 22,050 | "Put it there, it will, the AI will start finding out, okay, Dr. Munish Jindal can be a potential co(...TRUNCATED) | 7.4 | neutral | conversational | false | false | 0.75 |
en_src001_seg014 | en-IN | [0.000518798828125,0.000701904296875,0.000701904296875,0.000457763671875,0.00018310546875,0.00009155(...TRUNCATED) | 22,050 | "It will ask you whether you want to connect with this person or not, will share his LinkedIn profil(...TRUNCATED) | 7.1 | excited | formal | false | true | 0.65 |
en_src001_seg015 | en-IN | [-0.000732421875,-0.000579833984375,-0.00042724609375,-0.00054931640625,-0.00067138671875,-0.0007019(...TRUNCATED) | 22,050 | "It will go to Dr. Munish Jindal, will share your LinkedIn profile with him and ask his permission t(...TRUNCATED) | 8.800045 | neutral | informational | false | false | 0.55 |
en_src001_seg017 | en-IN | [-0.00006103515625,-0.00030517578125,0.000518798828125,0.001251220703125,0.0009765625,0.000427246093(...TRUNCATED) | 22,050 | "So this is what will happen during the lunch time and you can do it entire day. The bot is also giv(...TRUNCATED) | 9.6 | excited | formal | false | true | 0.65 |
End of preview. Expand in Data Studio
Sarvam TTS Dataset: Indian English and Hindi
A curated speech dataset for TTS model training, containing 396 clips in Indian English (en-IN) and Hindi (hi-IN) totalling 57.3 minutes (3440 seconds).
Pipeline source code: https://github.com/Ayush147258/sarvam-tts-dataset
Dataset Summary
| Stat | Value |
|---|---|
| Total clips | 396 |
| Total duration | 57.3 minutes (3440 seconds) |
| Clip duration range | 5.0s to 26.5s |
| Mean clip duration | 8.7s |
| Median clip duration | 7.5s |
| Manually verified | 0 / 396 (0%) |
| Automatically tagged only | 396 |
Language Balance
| Language | Clips | % of total |
|---|---|---|
| en-IN | 253 | 63.9% |
| hi-IN | 143 | 36.1% |
Emotion and Style Tags
| Emotion tag | Clips | % of total |
|---|---|---|
| neutral | 209 | 52.8% |
| happy | 61 | 15.4% |
| motivational | 47 | 11.9% |
| excited | 26 | 6.6% |
| sad | 19 | 4.8% |
| formal | 17 | 4.3% |
| angry | 15 | 3.8% |
| conversational | 1 | 0.3% |
| informational | 1 | 0.3% |
Valid emotion tags include neutral, happy, sad, excited, angry, formal, informational, storytelling, conversational, motivational, whisper, dramatic, and uncertain.
Manual Verification
Human audit status is stored in the manually_verified column. Rows with manually_verified: false use automatic ASR and automatic emotion/style tags and should be treated with lower confidence.
Dataset Schema
| Column | Description |
|---|---|
id |
Unique clip identifier |
language |
en-IN or hi-IN |
audio |
WAV audio decoded by datasets.Audio |
transcript |
Manual correction or ASR transcript |
duration_sec |
Clip duration in seconds |
emotion |
Emotion tag |
style |
Delivery style tag |
speaker_gender |
male, female, or unknown |
source_url |
Original source URL |
source_title |
Source title |
snr_db |
Estimated SNR |
sample_rate |
Expected sample rate |
quality_score |
Composite quality score from 0 to 1 |
manually_verified |
Whether a human reviewed transcript and tags |
created_at |
Processing date |
Usage
from datasets import load_dataset
ds = load_dataset("ayush712145/sarvam-tts-dataset")
row = ds["train"][0]
print(row["transcript"])
print(row["audio"]["sampling_rate"])
Limitations
Emotion and style tags are derived primarily from transcripts, so prosody-only cues such as sarcasm or subdued excitement may need manual correction. Source licensing must be verified before public redistribution.
- Downloads last month
- 22