Answers are grounded in this paper and cite their sources.
Demonstrates a new speech encoding method achieves 80% intelligibility at 2400 bps, suggesting efficiency.
Developing a speech encoding technique that requires less than 4000 bits per second (bps) has been considered challenging. Previous methods were based on signal prediction, taking the auditory functions into account but overlooking features specific to speech signals. A new method, called 1-bit vocoding, was developed in which amplitude envelopes in four frequency bands corresponding to spectral factors common to different languages were used to modulate infinitely peak-clipped signals, which also had been revealed to contain useful linguistic information. In a pilot experiment, intelligibility reached ∼80% with limited information of only 2400 bps, whereas the bit rate of the original signal was 256 000 bps. Noise-vocoding may require an even smaller bit rate; however, such severely degraded speech usually requires listeners for extensive training to be intelligible. Our algorithm preserves the naturalness of speech and is easy to grasp intuitively, while requiring an extremely low bit rate.
No takes yet. Share an insight, caveat, or question.
Ueda et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: