Skip to main content
SplitOptions controls how Tafrigh splits your audio files into manageable chunks. This is crucial for working with services like Wit.ai that impose duration limits. The splitting strategy uses silence detection to avoid cutting through words, which would reduce transcription accuracy.

​
Why splitting matters

Splitting audio files into chunks serves several purposes:
  • API compatibility: Services like Wit.ai have maximum duration limits per request
  • Parallel processing: Multiple chunks can be transcribed concurrently using different API keys
  • Quality preservation: Splitting on silence prevents cutting through words
  • Timestamp granularity: Chunk duration affects the detail level of your timestamps

​
Configuration

You pass splitOptions to the transcribe() function: 
import { transcribe } from 'tafrigh';

const transcript = await transcribe('audio.mp3', {
  splitOptions: {
    chunkDuration: 60,
    chunkMinThreshold: 4,
    silenceDetection: {
      silenceThreshold: -30,
      silenceDuration: 0.5,
    },
  },
});

​
Properties

​
chunkDuration
number
default:"60"
Maximum length of each audio chunk in seconds.The actual chunk length may be shorter if splitting at the maximum duration would cut through speech. Tafrigh automatically finds the last silence before the duration limit to ensure clean splits.
Must be between 4 and 300 seconds. This value also affects timestamp granularity—longer chunks mean less detailed timestamps.
​
chunkMinThreshold
number
default:"0.9"
Minimum length of each chunk in seconds.Chunks shorter than this threshold are automatically filtered out. This prevents creating tiny, unusable chunks from brief audio artifacts.
​
silenceDetection
object
Configuration for detecting silence in the audio to determine optimal split points.

​
Examples

​
Fast-paced podcast

For rapid conversation with minimal pauses: 
const transcript = await transcribe('podcast.mp3', {
  splitOptions: {
    chunkDuration: 45,
    chunkMinThreshold: 2,
    silenceDetection: {
      silenceThreshold: -25,
      silenceDuration: 0.2,
    },
  },
});

​
Lecture with long pauses

For educational content with deliberate pauses: 
const transcript = await transcribe('lecture.mp3', {
  splitOptions: {
    chunkDuration: 90,
    chunkMinThreshold: 5,
    silenceDetection: {
      silenceThreshold: -30,
      silenceDuration: 0.8,
    },
  },
});

​
Noisy environment recording

For audio with significant background noise: 
const transcript = await transcribe('interview.mp3', {
  splitOptions: {
    chunkDuration: 60,
    chunkMinThreshold: 3,
    silenceDetection: {
      silenceThreshold: -20, // Less sensitive to noise
      silenceDuration: 0.4,
    },
  },
});

​
Maximum duration chunks

For the longest allowed chunks (better for slower processing or API limits): 
import { transcribe, MAX_CHUNK_DURATION } from 'tafrigh';

const transcript = await transcribe('long-file.mp3', {
  splitOptions: {
    chunkDuration: MAX_CHUNK_DURATION, // 300 seconds
  },
});

​
How chunks are processed

After splitting, each chunk undergoes additional processing:
  1. Padding: Silent padding is added to chunk boundaries
  2. Normalization: Volume levels are normalized for consistency
  3. Filtering: Chunks below chunkMinThreshold are removed
This ensures optimal transcription quality even at split boundaries.
If you split in the middle of a word, transcription accuracy will suffer. Spend time tuning silenceDetection parameters for your specific audio content.

Preprocess options

Configure noise reduction and audio preprocessing

Transcribe function

Main transcription function reference

Build docs developers (and LLMs) love

Get started for freeTalk to us