Real-time acquisition and processing of speech signals

< a > Visual design of Speech signal acquisition and analysis system based on MATLAB

Abstract: This paper designs and develops a visual system of speech acquisition and analysis based on MATLAB, which realizes real-time acquisition of voice signal by Realtek AC97 sound card and MATLAB data collection toolbox, and uses MATLAB The powerful numerical calculation and signal processing function completes the speech signal analysis work with high accuracy. The system is also optimized using MATLAB's graphical user interface design tool, which replaces complicated program modification and debugging process with simple dialog box and menu operation, which makes the system more convenient and flexible.

The thesis understands: The system mainly includes the voice acquisition and the processing two parts, uses the MATLAB data collection toolbox through the notebook own sound card to collect the external analog voice signal, the quantization obtains the digital voice signal, the collected data temporarily saves in the memory or the disk, Then using MATLAB to process and analyze the corresponding arithmetic of digital speech signal, finally, the audio signal is converted into analog voice signals and sent to the speaker to play. This design is suitable for the simulation of the algorithm, the sound card in the paper has the highest 16-bit sampling number and 44.1kHz sampling rate, can be a high level of human voice fidelity acquisition.

MATLAB has three ways to drive the sound card: the establishment of analog input device objects, Wavrecord, Audiorecorder.

The main characteristic parameters of speech signal: short time energy, short time average amplitude, short time average over 0 rate.

For the real-time processing to be further studied, in the computer is not possible to achieve full real-time, the collected data is generally saved to the data buffer, and then through the algorithm processing, the final playback, in the process of processing will inevitably have a delay time. There are two questions to consider, first, how to set the size of the data buffer to ensure that the data collected will be processed without being overwritten, second, how to ensure that after the algorithm processing of the sound is maintained continuously, which will require the algorithm processing time less than the data buffer to save the playback time of the signal.

< two > real-time voice acquisition and processing system based on DSP

Abstract: This paper discusses the real-time digital sound processing system based on dsp56f862 EVM development system, in the reverberation of digital audio signal , based on the research and simulation of the theory and algorithm of the synthesis of chorus, equalization and other sound methods, the processing of these sound effects is realized on the hardware. and can be played after real-time processing.

The thesis understands: The system mainly consists of dsp56f826 and CodecCS4218, the person's song is converted to the level signal by the mic, the codec A/D conversion to the digital signal is saved to the input data buffer, the dsp56f826 calls the algorithm to data processing and output to the output buffer, Finally, the codec D/a conversion is converted to analog signal and output by the amplifier circuit. This design scheme is suitable for the actual design, the different scheme uses the concrete algorithm to carry on the analysis, has the practical application significance. For people's song processing, and then to form a better effect, the key is the specific algorithm design.

The dsp56f826 operation speed is 40MIps, the 64kx16bit program store, the 64kx16bit data store, and the 70MHz operation can wait for access to the external memory 0.

The codec includes A/d, D/A converter, 16-bit stereo, using 12.288M crystal sampling frequency of 8~48khz.

　　DSP and codec through the SSI serial communication, codec each time to send a 16bit of sampling data to the DSP, the DSP to save the received data to the input buffer, while the stored data processed, and then deposited in the output buffer, The SSI output interrupt program of the DSP is timed to execute from the output buffer, and finally sent to the codec to output in a simulated manner, thus processing in real time.

Some of the results of the algorithm implementation:

Reverb Effect: Reverb sound must occur after a period of time in the original singing, the human ear to the time interval of 30~50ms above the two signals to distinguish out. (IIR filter)

Chorus effect: Like a reverb effect, the difference is that its delay time function is a low-frequency signal that changes over time. (Maximum number of people, more will be added to change the tone, sound frequency and other time-varying functions)

Equalization Effect: the signal in a certain frequency band is enhanced or attenuated to improve the output frequency response characteristics, improve the auditory effect.

So how do you achieve the effects of vocals or harmonies? I think in the original singing sound on the basis of harmony with a relatively high feasibility, first, with the synthesis of the sound of the original singing is realistic, the second, the accompaniment is a person singing, lower or higher than the original singing tone, then the original singing to the appropriate processing to produce a harmony effect will be better. Although it is similar to the effect of chorus or Echo, but the key is when to need harmony, when do not need harmony? Do you want to judge the original singing in theory, such as the main melody, or the climax part? But for real-time systems, this is hard to define.

Real-time acquisition and processing of speech signals