Spontaneous IVR

Our offering of speech data comprised of participants speaking a query regarding a certain topic in their own words and repeating that query twice more per recording in different words. The recordings are saved in dual channel 8Khz 16 bit over telephony, with the second channel containing the TTS prompts. Like our H2H dialogue offering, the participant’s channel is then transcribed and validated for the lowest possible word error rate. The resulting dataset is a good start for those looking to create telephony-robust models that have a degree of spontaneous speech, like call center IVR applications.