ErfanRou/callcc-test-1k-samples-soniox

ErfanRou/callcc-test-1k Full-channel benchmark set for Persian call-centre ASR: one row per channel of a call (0 = agent, 1 = customer), with the complete 16 kHz mono channel audio and the complete Soniox stt-async-v5 transcript of that channel, rebuilt from the raw tokens of ErfanRou/callcc-test-1k-windowed (no re-transcription). Use it to evaluate the serving path (whole-channel input, model-side VAD/chunking) with corpus-level WER/CER — see eval_full_call.py in the kit. text… See the full description on the dataset page: https://huggingface.co/datasets/ErfanRou/callcc-test-1k-samples-soniox.