File size: 1,314 Bytes
8183a8d
9f22f1a
 
 
 
 
 
 
 
 
 
8183a8d
9f22f1a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aee81ec
 
 
 
 
307b505
aee81ec
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
---
base_model: bulbul:v3
language:
- hi
- en
pretty_name: Hinglish Audio Turn-End Example
tags:
- audio
- hinglish
- speech
- text-to-speech
---

# Hinglish Audio Turn End Example

Dataset of hinglish audio snippets generated with SarvamAI `bulbul:v3` TTS.

Each example is a customer service utterance with metadata about the speaker
gender, whether the utterance is complete or incomplete (trailoff / connective /
filler ending), and the rendered audio file.

## Fields

| field | type | description |
|---|---|---|
| `script` | string | hinglish transcript (Devanagari + Latin script mixed) |
| `audio` | audio | mp3 speech clip generated by SarvamAI TTS |
| `gbvc` | string | gender/voice class: `male`, `female`, or `neutral` |
| `ending` | string | utterance ending type: connector, filler, trailoff, or empty (complete) |
| `completeness` | string | `complete` when `ending` is empty, else `incomplete` |

## Speaker mapping

| `gbvc` | speaker |
|---|---|
| `male` | `shubh` |
| `neutral` | `shubh` |
| `female` | `ritu` |

## TTS settings

- model: `bulbul:v3`
- language: `hi-IN`
- pace: `1`
- sample rate: 16000 Hz

## Aproximate cost for this dataset

- Script generation model: deepseek-v4-flash-latest:
    - Cost: $0.1
    - Uncached Tokens: 481k, Cached Tokens: 3.12m
- TTS Cost: apprx 65 ruppees