Speech Synthesis Algorithm Research And FPGA Implementation

With the development of science and technology, human-computer interaction technology has become one of hot-researching fields. The speech synthesis is an important part of the technology, and is widely used. With the embedded hardware's performance improved, the storage capacity increased and the price decreasing, embedded speech synthesis technology is being widely used in some applications. FPGA has become the most popular embedded platform nowdays for the flexible and extensive characteristic.This paper compares the advantages and disadvantages of all kinds of speech synthesis algorithms, and chooses Time-domain Pitch-synchronous Overlap and Add (TD-PSOLA) as the speech synthesis module algorithm. Using MATLAB software, the TD-PSOLA algorithm is designed and analysed, optimizated, validated. The SOPC system is establishabled by Quartus II software. Then in the Nios II IDE environment, TD-PSOLA algorithm is writed and run.And the system accomplishes the function of speech synthesis.The experiment proves that the TD-PSOLA algorithm accomplishes the regulation of speech rhythm well in a certain range. Pitch detection accuracy comes up to 89.2% with the peak method and the best path method. And by using energy threshold and zero-rate threshold adjustment, the voicingless/voicing determination accuracy reaches 93.16%. The designed SOPC platform can satisfy systerm needs, and provide bigger space for improvement.Listening test results show that the change of speech rhythm is limited based on TD-PSOLA-based speech synthesis algorithm. When the long-range changes from 0.5 to 3 times and the range of pitch changes is at 0.5 to 2.5 times, speech synthesis results are satisfactory. Multi-level design of voice samples can make smaller modification of speech synthesis,and improve synthesized speech quality.