Due to Halef originating from a conventional telephony platform, it is currently limited to 8kHz sampling rate for audio in- and output. This is suboptimal for ASR as well as perceived TTS quality. Explore how to make the sampling rate generic. This is particularly interesting when using softphones, web clients, or smart phone apps as SIP clients. Potentially affected modules include
Dear David,
The DNN-AM training procedure has been finished successfully.
Please see the attached file for the results.
Best!
AI
Dear Alex,
This is great! A mere 0.93% WER difference between 8kHz and 16kHz speech is gorgeous and shows that adding enhanced sampling rate support in Halef is of rather low priority compared to other activities.
Thanks,
Yours,
DSO