CN103117058B - Based on Multi-voice engine switch system and the method for intelligent television platform - Google Patents

Abstract

The invention relates to a smart TV software platform, which discloses a multi-speech engine switching method based on a smart TV platform, which realizes automatic search and switching of the speech engine with the highest current recognition efficiency, and improves the user's speech interaction experience. The method can be summarized as follows: when the user runs the voice application program to use the voice recognition function, the voice engine selection module obtains the collected voice data through the voice application interface, and then sends the voice data to each voice engine module, records and compares each voice The engine module returns the response time of the recognition result, and the speech engine module with the shortest response time is selected for switching. In addition, the invention also discloses a corresponding switching system, which is suitable for realizing the fast speech recognition function in the smart TV.

Description

Multi-speech engine switching system and method based on smart TV platform

technical field

The invention relates to a smart TV software platform, in particular to a multi-voice engine switching system and method based on a smart TV platform.

Background technique

With the development of intelligent and networked TV terminals, the content available to smart TVs has been greatly enriched, and the functions have become more diversified, and the control of TVs has become more frequent and complicated. The application of speech recognition technology on smart TVs greatly simplifies the user's operation process, and the user experience is greatly improved. Since speech recognition needs to occupy huge system resources, smart TVs are generally connected to cloud servers through the network to realize speech recognition functions;

The voice recognition engine used in the server to realize the voice recognition function is composed of a voice detection module, a feature extraction module and a recognition search module; wherein, the function of the voice detection module is to detect and process voice signals, and the TV will collect the original The voice data is sent to the module, and the voice signal data needs to be converted into a standard data format (for example: 8K, 16bit) in the voice detection module; at the same time, the start and end points of the voice are judged by using an efficient signal detection algorithm; The feature extraction module receives the detected voice data stream, and extracts the feature vector stream of the voice signal from it. Speech feature is the use of digital signal processing technology to extract information that best reflects its essential attributes from the speech signal. In this module, it is necessary to perform pre-emphasis, framing, windowing, product and transformation, cepstrum transformation, and difference processing on the speech signal, and finally obtain a feature vector of about tens of dimensions; the unknown speech that the recognition search module will receive The signal features are matched with the acoustic model library, dictionary/dictionary and recognition grammar information in the engine to obtain the word sequence most suitable for unknown speech features. This process can be briefly described as follows: By retrieving a dictionary/dictionary, a sentence can be decomposed from a sequence of words into a sequence of phonemes. This phoneme sequence is combined with the acoustic model to obtain acoustic model unit sequence information that reflects its essential properties. Then, the feature vector of the original speech is matched with the information of all possible sentence candidate acoustic model unit sequences, the matching probability is calculated, and the acoustic model unit sequence with the largest posterior probability is selected. Through this unit sequence, the corresponding word sequence can be obtained, which is the text sequence output by the engine to the TV.

And because there are multiple speech recognition engines in the server, if a single fixed engine is used for speech recognition, it is not conducive to the improvement of the efficiency of smart TV speech recognition, resulting in poor user experience in speech interaction; therefore, how to use multiple speech recognition engines Finding the current most efficient speech recognition engine and switching between them is an urgent problem to be solved in speech interaction applications.

Contents of the invention

The technical problem to be solved by the present invention is to propose a multi-speech engine switching system and method based on a smart TV platform, realize automatic search for and switch the speech engine with the highest current recognition efficiency, and improve the user's speech interaction experience.

The solution adopted by the present invention to solve the above-mentioned technical problems is: a multi-voice engine switching system based on the smart TV platform, comprising: a voice engine selection module and at least two voice engine modules; all voice engine modules are encapsulated by a unified voice engine interface, And the voice engine selection module is connected through the voice engine interface; the voice engine selection module is connected with the voice application program through the voice application interface.

Further, the voice engine module is used to obtain the voice data transmitted by the voice engine selection module from the voice engine interface, and recognize the voice data, and then return the recognition result to the voice engine selection module; When the application program uses the voice recognition function, it obtains the collected voice data through the voice application interface, sends the voice data to each voice engine module through the voice engine interface, and receives the recognition results returned by all voice engine modules, and records each voice engine module The response time of the recognition result is returned and compared, and the speech engine module with the shortest response time is selected for switching, so that the speech application program can call the speech engine module with the highest recognition efficiency.

Further, the selection of the speech engine module with the shortest response time for switching refers to: the speech engine selection module is connected to the speech engine module with the shortest response time through the speech engine interface, and disconnected from other speech engine modules at the same time.

In addition, the present invention also proposes a corresponding smart TV platform-based multi-speech engine switching method, including:

a. When the user runs the voice application program to use the voice recognition function, the voice engine selection module obtains the collected voice data through the voice application interface;

b. the voice engine selection module sends the voice data to each voice engine module through the voice engine interface;

c. each voice engine module identifies the voice data, and then returns the recognition result to the voice engine selection module;

d. The speech engine selection module records and compares the response time of each speech engine module returning the recognition result, and selects the speech engine module with the shortest response time to switch.

Further, in step d, selecting the speech engine module with the shortest response time to switch means: the speech engine selection module is connected to the speech engine module with the shortest response time through the speech engine interface, and disconnected from other speech engine modules at the same time.

The beneficial effects of the present invention are: by comparing the response time (i.e. recognition speed) of the recognition results returned by each speech engine module, the speech engine module with the shortest response time is selected for switching, so that the speech application program can call the speech with the highest recognition efficiency The engine module performs speech recognition, thereby improving the overall recognition efficiency of speech recognition; and, since the connection carrier (voice application interface) between the speech application program and the speech engine selection module remains unchanged, when the speech engine module switches, the speech The application does not need to pay attention to which speech engine module is switched, thus ensuring the stability and continuity of speech recognition.

Description of drawings

Fig. 1 is based on the multi-speech engine switching system of intelligent TV platform among the present invention and realizes frame diagram;

Fig. 2 is a flow chart of the multi-speech engine switching method based on the smart TV platform in the present invention.

Detailed ways

The realization principle of the present invention is: due to the performance difference of each speech engine module in the system, these modules just have fast or slow to the processing of speech data, therefore, we can process speech to each speech engine module by arranging a speech engine selection module The response time of the data is recorded and compared, so as to find out the voice engine module with the shortest processing time and the fastest response, and then switch to the connection of the voice engine module, and the introduction of the voice engine selection module is due to its relationship with the voice application program The application interface among them has not changed all the time, therefore, it can also solve the stability problem of the system at the same time.

Referring to Fig. 1, among the present invention, the multi-speech engine switching system based on smart TV platform comprises speech engine selection module and a plurality of speech engine modules; All speech engine modules are encapsulated by unified speech engine interface, and connect voice Engine selection module; the speech engine selection module is connected with the speech application program through the speech application interface.

Wherein, the voice engine module is used to obtain the voice data transmitted by the voice engine selection module from the voice engine interface, and recognizes the voice data, and then returns the recognition result to the voice engine selection module; When the application program uses the voice recognition function, it obtains the collected voice data through the voice application interface, sends the voice data to each voice engine module through the voice engine interface, and receives the recognition results returned by all voice engine modules, and records each voice engine module The response time of the recognition result is returned and compared, and the speech engine module with the shortest response time is selected for switching, so that the speech application program can call the speech engine module with the highest recognition efficiency.

Figure 2 shows the corresponding flow of the handover method, which includes the following implementation steps:

a. When the user runs the speech application program to use the speech recognition function, the speech engine selection module obtains the collected speech data through the speech application interface; the speech data comes from the sound source signal collected by the speech collection device of the smart TV;

b. The voice engine selection module sends the voice data to each voice engine module through the voice engine interface; due to the adoption of a unified voice engine interface for encapsulation, each voice engine module can receive the same voice data at the same time;

c. each voice engine module identifies the voice data, and then returns the recognition result to the voice engine selection module;

d. the voice engine selection module records the response time of each voice engine module to return the recognition result and compares it, selects the voice engine module with the shortest response time to switch: the voice engine selection module is connected to the voice engine module with the shortest response time through the voice engine interface, Disconnect from other voice engine modules at the same time. Afterwards, voice applications can implement fast voice recognition by invoking the voice engine module with the shortest response time, thereby improving the user's voice interaction experience.

Claims (2)

1. The multi-voice engine switching system based on the smart TV platform is characterized in that, comprising: a voice engine selection module and at least two voice engine modules; all voice engine modules are encapsulated by a unified voice engine interface, and through the voice engine interface Connect the voice engine selection module; the voice engine selection module is connected with the voice application program through the voice application interface; The voice engine module is used to obtain the voice data transmitted by the voice engine selection module from the voice engine interface, and recognizes the voice data, and then returns the recognition result to the voice engine selection module; the voice engine selection module is used for voice application When using the speech recognition function, obtain the collected speech data through the speech application interface, send the speech data to each speech engine module through the speech engine interface, and receive the recognition results returned by all speech engine modules, and record the recognition returned by each speech engine module The response time of the results is compared, and the speech engine module with the shortest response time is selected for switching, so that the speech application program can call the speech engine module with the highest recognition efficiency; The selection of the speech engine module with the shortest response time for switching means that the speech engine selection module is connected to the speech engine module with the shortest response time through the speech engine interface, and disconnected from other speech engine modules at the same time. 2. based on the multi-speech engine switching method of smart TV platform, be applied in the system as claimed in claim 1, it is characterized in that, comprising: a. When the user runs the voice application program to use the voice recognition function, the voice engine selection module obtains the collected voice data through the voice application interface; b. the voice engine selection module sends the voice data to each voice engine module through the voice engine interface; c. each voice engine module identifies the voice data, and then returns the recognition result to the voice engine selection module; d. The speech engine selection module records the response time of each speech engine module to return the recognition result and compares it, and selects the speech engine module with the shortest response time to switch; In step d, selecting the speech engine module with the shortest response time to switch means: the speech engine selection module is connected to the speech engine module with the shortest response time through the speech engine interface, and disconnected from other speech engine modules at the same time.