Understanding preprocessing
Tafrigh automatically preprocesses audio files before transcription to improve accuracy. You can customize noise reduction, filtering, and dialogue enhancement to match your audio characteristics.
Default preprocessing
By default, Tafrigh applies noise reduction with these settings:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
// Uses default preprocessing
const transcript = await transcribe('audio.mp3');
- High-pass filter at 300 Hz
- Low-pass filter at 3000 Hz
- FFT-based denoising from 0s to 1.5s
- Noise floor at -20 dB
- Dialogue enhancement enabled
Custom noise reduction
Adjust noise reduction parameters to match your audio environment:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('noisy-audio.mp3', {
preprocessOptions: {
noiseReduction: {
// Isolate voice frequencies (remove low-frequency rumble)
highpass: 200,
// Remove high-frequency hiss
lowpass: 3000,
// FFT-based noise reduction timing
afftdnStart: 1,
afftdnStop: 2,
// Noise floor threshold
afftdn_nf: -25,
// Enhance dialogue clarity
dialogueEnhance: true,
},
},
});
console.log(transcript);
Parameter guide:
highpass - Removes low frequencies below this value (Hz). Useful for eliminating rumble and bass noise.lowpass - Removes high frequencies above this value (Hz). Useful for eliminating hiss and electronic noise.afftdnStart - When to start analyzing noise profile (seconds)afftdnStop - When to stop analyzing noise profile (seconds)afftdn_nf - Noise floor in dB. Lower values = more aggressive noise reductiondialogueEnhance - Boosts midrange frequencies where speech occurs
Disabling noise reduction
For clean studio recordings, you may want to skip noise reduction:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('studio-recording.mp3', {
preprocessOptions: {
noiseReduction: null,
},
});
console.log(transcript);
Disabling noise reduction speeds up processing for high-quality audio that doesn’t need enhancement.
Selective filter usage
You can disable individual filters by setting them to null:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('audio.mp3', {
preprocessOptions: {
noiseReduction: {
// Only use high-pass filter, disable others
highpass: 300,
lowpass: null,
afftdnStart: null,
afftdnStop: null,
afftdn_nf: null,
dialogueEnhance: false,
},
},
});
console.log(transcript);
Aggressive noise reduction
For very noisy environments (street recordings, crowded spaces):
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('street-interview.mp3', {
preprocessOptions: {
noiseReduction: {
highpass: 250,
lowpass: 2500,
afftdnStart: 0.5,
afftdnStop: 3,
afftdn_nf: -30, // More aggressive
dialogueEnhance: true,
},
},
});
console.log(transcript);
Phone call quality audio
Optimize for telephone or VoIP recordings:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('phone-call.mp3', {
preprocessOptions: {
noiseReduction: {
// Telephone bandwidth is typically 300-3400 Hz
highpass: 300,
lowpass: 3400,
afftdnStart: 0,
afftdnStop: 1.5,
afftdn_nf: -20,
dialogueEnhance: true,
},
},
});
console.log(transcript);
Monitoring preprocessing progress
Track preprocessing progress with callbacks:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('audio.mp3', {
preprocessOptions: {
noiseReduction: {
dialogueEnhance: true,
highpass: 200,
lowpass: 3000,
},
},
callbacks: {
onPreprocessingStarted: async (filePath) => {
console.log(`Starting preprocessing: ${filePath}`);
},
onPreprocessingProgress: async (percent) => {
console.log(`Preprocessing progress: ${percent}%`);
},
onPreprocessingFinished: async (filePath) => {
console.log(`Preprocessing complete: ${filePath}`);
},
},
});
console.log(transcript);
Expected output
Starting preprocessing: /tmp/tafrigh/1234567890.mp3
Preprocessing progress: 15%
Preprocessing progress: 35%
Preprocessing progress: 58%
Preprocessing progress: 82%
Preprocessing progress: 100%
Preprocessing complete: /tmp/tafrigh/1234567890.mp3
Custom splitting options
Combine preprocessing with custom audio splitting for optimal results:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['your-wit-ai-key'] });
const transcript = await transcribe('long-lecture.mp3', {
preprocessOptions: {
noiseReduction: {
dialogueEnhance: true,
highpass: 250,
lowpass: 3000,
},
},
splitOptions: {
// 60-second chunks
chunkDuration: 60,
// Minimum chunk length (filter out very short segments)
chunkMinThreshold: 4,
silenceDetection: {
// What volume level counts as silence
silenceThreshold: -30,
// How long silence must last to trigger a split
silenceDuration: 0.5,
},
},
});
console.log(transcript);
Adjusting silenceThreshold is important when you have background noise. A lower value (e.g., -40 dB) is stricter and works for quiet environments, while a higher value (e.g., -20 dB) is more lenient for noisy recordings.
Podcast optimization
Recommended settings for podcast transcription:
import { init, transcribe } from 'tafrigh';
init({ apiKeys: ['key1', 'key2', 'key3'] });
const transcript = await transcribe('podcast-episode.mp3', {
concurrency: 3,
preprocessOptions: {
noiseReduction: {
highpass: 100, // Keep some bass for voice warmth
lowpass: 3500,
afftdnStart: 0,
afftdnStop: 1,
afftdn_nf: -15, // Light noise reduction
dialogueEnhance: true,
},
},
splitOptions: {
chunkDuration: 60,
chunkMinThreshold: 2,
silenceDetection: {
silenceThreshold: -35,
silenceDuration: 0.8, // Longer pauses between speakers
},
},
});
console.log(transcript);
For music podcasts or content with intentional audio effects, use lighter noise reduction settings or disable it entirely to preserve audio quality.