INA Speech Segmenter (--detector ina)

Uses a CNN (Convolutional Neural Network) trained on broadcast media to detect speech vs music vs noise vs silence.

Install

pip install "praisonai-editor[detect]"

Usage

praisonai-editor edit file.mp3 --preset speech_only --detector ina

Strengths

• High accuracy for speech vs music boundary detection
• Handles noisy environments well (broadcast quality)
• Good at detecting short speech segments within music

Weaknesses

• Slower (CNN inference)
• Cannot distinguish singing from music (both labeled as music)
• Use with --demix to get singing classification

Content types returned

INA label	Mapped to
speech	speech
music	music
noEnergy	silence