Norwegian 91.73h ASR / LLM / STT / TTS Training dataset, HUMAN transcribed

This dataset contains a broad set of spontaneous dual channel dialogues, dataset was created specifically for ASR STT / TTS training. It covers 91.73 hours of native Norwegian speech, delivered in dual channel (channel per speaker, separated speakers) audio files and human transcribed (97% acc) in Bokmål Audio is already segmented and clipped, 67447 wav files (16khz), average clip length 4.9 seconds. Total duration (effective audio duration, total clip duration summed) 91.73 hours / 91 hours and 44 minutes. Segmentation and Transcription is 100% HUMAN MADE Dataset has 68 unique speakers, 36 males, 32 females. Source audio was dual channel, so there is ZERO overlap of speakers, clear one speaker audio on every clip. Dataset contains: - 67447 16khz wav files, already volume normalized - Average clip / wav file length 4.9 seconds - Total duration (effective clip duration summed, all silent parts between speakers already removed) 91 hours and 44 minutes - 68 unique speakers, 36 males, 32 females - Human segmented (clipped) and transcribed with 97% accuracy SAMPLE of 40 clips with transcription json: https://drive.google.com/file/d/14Zp3rY-Jk8u8aqAJVl4jMCjTQIrjFxEd/view?usp=sharing