Coding method and device for speech synthesis

Abstract

One or more embodiments of the invention provide a coding method and device for speech synthesis. According to the coding method, after text data is synthesized into pcm stream data, the initial buffer threshold T(start) required for starting playing is dynamically calculated according to a current system load condition, and if the buffer duration of a buffer area is greater than the initial buffer threshold T(start), the pcm stream data in the buffer area is read and played; a continuous buffer threshold T(block) required by continuous playing is dynamically calculated according to the current system load condition; and in the process of playing, whether the text data continues to be synthesized into the pcm stream data or synthesis is paused is judged according to a relation between the buffering duration of the buffer area and the continuous buffer threshold T(block), so the stability and smoothness of playing are guaranteed, and smooth occupation of the cpu and the memory is also achieved.

Description

technical field [0001] The present invention relates to the technical field of speech synthesis methods, in particular to a speech synthesis encoding method and device. Background technique [0002] The current speech synthesis (text-to-speech) encoding playback schemes are: [0003] 1. One-time synthesis: input the text into the speech synthesis engine, obtain the encoded pcm data at one time, and pass the pcm data to the player for playback at one time; this method requires a large amount of memory to store pcm, and the waiting time for synthesis is longer , you need to wait until all the data is synthesized before starting to play. [0004] 2. Sleep during streaming synthesis: While synthesizing pcm data, stuff it into the player for processing, sleep for a certain period of time during the synthesis process, continue to synthesize and play; the data block size of a single synthesis in this method is fixed, there are cpu fluctuations, sleep time If the sleep time is too...

AI Technical Summary

Problems solved by technology

[0003] 1. One-time synthesis: input the text into the speech synthesis engine, obtain the encoded pcm data at one time, and pass the pcm data to the player for playback at one time; this method requires a large amount of memory to store pcm, and the waiting time for synthesis is longer , you need to wait for all the data to be synthesized before starting to play

[0004] 2. Sleep during streaming synthesis: While synthesizing pcm data, stuff it into the player for processing, sleep for a certain period of time during the synthesis process, continue to synthesize and play; the data block size of a single synthesis in this method is fixed, there are cpu fluctuations, sleep time If the sleep time is too short, the cpu may be too high. If the sleep time is too long, the player may stop broadcasting or cause noise.

Images