JP2015161718A - speech detection device, speech detection method and speech detection program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speech detection device capable of easily detecting a voice question to a system without receiving any special operation, a speech detection method and a speech detection program.SOLUTION: A speech detection device 1 includes: a voice acquisition part 11 for continuously acquiring voice data; a time measuring part 12 for measuring a silence time t1 in which a level of an acquired voice is lower than an utterance determination threshold value a1, a speech time t2 in which the level of the voice exceeds the utterance determination threshold value a1 after the lapse of the silence time t1, and a non-response time t3 in which the level of the voice is lower than the utterance determination threshold value a1 after the lapse of the speech time t2; and a determination part 13 by which, in the case where the silence time t1, the speech time t2 and the non-response time t3 meet a predetermined condition, the voices during the speech time t2 are determined as a speech of a question to the device.

Description

  The present invention relates to an apparatus, a method, and a program for detecting an utterance from audio data.

  2. Description of the Related Art Conventionally, some electronic devices such as a car navigation device and a portable terminal are equipped with a function for executing a predetermined process in accordance with a result of voice recognition. In these electronic devices, generally, a voice recognition function is activated in response to a user performing a predetermined operation such as pressing a button.

  However, when the voice recognition function is activated in response to a user operation, for example, when both hands of the user are closed or when an instruction is given to a device that is difficult to perform a direct operation, the convenience is reduced. Therefore, for example, there are a method for determining a question to the system using feature values such as prosody in speech data (see Non-Patent Document 1) and a method for determining a conversation state from the length of a silent section (see Patent Document 1). Proposed.

JP 2005-196025 A

Tomoyuki Yamagata, Satoshi Sako, Tetsuya Takiguchi, Yasuo Ariki, “System Request Detection Using Prosody and Speaker Change Information”, 9th Spoken Language Symposium, SIG-SLP69, pp. 289-294, 2007-12

  However, processing with a high load using audio data feature amounts requires processing performance and power consumption of the CPU, which may be inconvenient for electronic devices such as portable terminals. Also, when determining based on the length of the silent section, it is difficult to determine only the question to the system.

  It is an object of the present invention to provide an utterance detection device, an utterance detection method, and an utterance detection program that can easily detect a voice inquiry to a system without receiving a special operation.

  The utterance detection device according to the present invention includes a voice acquisition unit that continuously acquires voice data, a first time when the size of the acquired voice does not satisfy the utterance determination threshold, and a voice after the first time has elapsed. A second time when the size of the voice exceeds the utterance determination threshold, and a time measuring unit that measures a third time when the voice size does not satisfy the utterance determination threshold after the second time has elapsed, A determination unit that determines, when the first time, the second time, and the third time satisfy a predetermined condition, that the voice at the second time is an inquiry utterance to the own device; Prepare.

  The predetermined condition is that the first time is greater than or equal to a first time threshold, the second time is within a range from a second time threshold to a third time threshold, and the third May include that the time is greater than or equal to a fourth time threshold.

  The utterance detection device includes an identification unit that identifies whether or not the pattern of the voice data in the second time is a human voice by a predetermined classifier, and the determination unit satisfies the predetermined condition And when the voice data pattern at the second time is identified as a human voice by the identification unit, the voice at the second time is determined to be an utterance of an inquiry to the own device. May be.

  The timekeeping unit may adjust the utterance determination threshold based on a statistical value related to the loudness of the first time and the third time.

  The utterance detection device may include an output unit that outputs a predetermined arithmetic processing result as a voice when it is determined that the voice at the second time is a questioned utterance to the own device.

  The speech detection device, when it is determined that the speech at the second time is an inquiry speech to the device itself, a speech analysis unit that executes a predetermined speech analysis process on the speech data acquired subsequently May be provided.

  The utterance detection device is stored in the storage unit when the voice data at the second time is stored, and when it is determined that the voice at the second time is an inquiry utterance to the own device. A voice analysis unit that performs a predetermined voice analysis process on the voice data.

  The speech detection method according to the present invention includes a voice acquisition step of continuously acquiring voice data, a first time when the size of the acquired voice does not satisfy the utterance determination threshold, and a voice after the first time has elapsed. Measuring a second time when the magnitude of the voice exceeds the utterance determination threshold, and a third time when the volume of the voice does not satisfy the utterance determination threshold after the second time has elapsed, and A determination step of determining, when the first time, the second time, and the third time satisfy a predetermined condition, that the voice at the second time is an inquiry utterance to the device; The computer runs.

  The speech detection program according to the present invention includes a speech acquisition step of continuously acquiring speech data, a first time when the size of the acquired speech is less than the speech determination threshold, and a speech after the first time has elapsed. Measuring a second time when the magnitude of the voice exceeds the utterance determination threshold, and a third time when the volume of the voice does not satisfy the utterance determination threshold after the second time has elapsed, and A determination step of determining, when the first time, the second time, and the third time satisfy a predetermined condition, that the voice at the second time is an inquiry utterance to the device; Let the computer run.

  According to the present invention, it is possible to easily detect a voice inquiry to the system without receiving a special operation.

It is a block diagram which shows the function structure of the speech detection apparatus which concerns on 1st Embodiment. It is a figure which shows an example of the audio | voice data which concern on 1st Embodiment. It is a flowchart which shows the flow of the speech detection process which concerns on 1st Embodiment. It is a block diagram which shows the function structure of the speech detection apparatus which concerns on 2nd Embodiment. It is a block diagram which shows the function structure of the speech detection apparatus which concerns on 3rd Embodiment. It is a block diagram which shows the function structure of the speech detection apparatus which concerns on 4th Embodiment.

[First Embodiment]
The first embodiment of the present invention will be described below.
The utterance detection device 1 of the present embodiment detects that the acquired voice data is an inquiry to the device itself.

FIG. 1 is a block diagram illustrating a functional configuration of the utterance detection apparatus 1.
The utterance detection device 1 includes a voice acquisition unit 11, a timer unit 12, a determination unit 13, and an output unit 14.
The voice acquisition unit 11 continuously acquires voice data.

FIG. 2 is a diagram illustrating an example of audio data acquired by the audio acquisition unit 11.
The sound acquisition unit 11 continuously acquires at least an amplitude indicating the size of the sound in time series.

  The timer 12 has a silence time t1 (first time) in which the acquired amplitude is less than the utterance determination threshold a1, a speech time t2 (second time) in which the amplitude exceeds a1 after t1 has elapsed, and t2 After the elapse of time, a non-response time t3 (third time) in which the amplitude is less than a1 is measured.

  The silence time t1 is an index for determining that the person is not in a conversation state. That is, the time measuring unit 12 measures the silence time t1 in accordance with the premise that a certain amount of silence occurs before an inquiry to the system (the utterance detection device 1) is uttered.

  The utterance time t2 is an index for determining that the voice is a question. That is, in the case of an inquiry to the system, the timer 12 measures the utterance time t2 in accordance with the premise that the utterance is shorter in time than the inquiry to the human.

  The no-response time t3 is an index for determining that the voice is an inquiry to the system and that no human responds. That is, the timer 12 measures the non-response time t3 in accordance with the premise that there is no response from a human being to a question to the system.

  In addition, the time measuring unit 12 adjusts the utterance determination threshold value a <b> 1 based on the statistic value of the loudness of the silence time t <b> 1 and the no-response time t <b> 3, that is, the magnitude of ambient noise when there is no utterance. Specifically, the time measuring unit 12 calculates the environmental sound level and the inter-individual error, for example, by calculating the standard deviation of the sound data at the silence time t1 and the no-response time t3, and speaks based on the calculation result. The determination threshold value a1 is adjusted. Thereby, the time measuring unit 12 increases the utterance determination threshold value a1 as the surrounding noise increases, and decreases the utterance determination threshold value a1 as the surrounding noise decreases, so that the utterance time t2 can be correctly measured.

When the silence time t1, the utterance time t2, and the no-response time t3 satisfy predetermined conditions, the determination unit 13 determines that the voice at the utterance time t2 is an utterance that is an inquiry to the utterance detection device 1.
The predetermined condition is that the silence time t1 is not less than a first time threshold (for example, 30 seconds), and the speech time t2 is from a second time threshold (for example, 0.5 seconds) to a third time threshold (for example, 0.5 seconds). 2 seconds), and the no-response time t3 is equal to or greater than a fourth time threshold (for example, 4 seconds).

  When it is determined that the voice at the utterance time t <b> 2 is an inquiry utterance to the utterance detection device 1, the output unit 14 outputs a result of a predetermined calculation process as a response. The output unit 14 outputs a predetermined response in response to an inquiry from the user, for example, acquiring the current time and reading it out.

FIG. 3 is a flowchart showing the flow of speech detection processing by the speech detection device 1.
This process is executed in parallel with the acquisition of audio data by the audio acquisition unit 11.

In step S1, the time measuring unit 12 starts measuring the silence time t1.
In step S2, the determination unit 13 determines whether the amplitude of the audio data exceeds the utterance determination threshold value a1. If this determination is YES, since the utterance has been started, the process proceeds to step S3. On the other hand, if the determination is NO, silence continues, so the determination in step S2 is repeated.
In step S3, the time measuring unit 12 determines the silence time t1.

In step S4, the time measuring unit 12 starts measuring the utterance time t2.
In step S5, the determination unit 13 determines whether the amplitude of the audio data exceeds the utterance determination threshold value a1. If this determination is YES, since the utterance continues, the determination in step S5 is repeated. On the other hand, if the determination is NO, since the utterance has ended, the process proceeds to step S6.
In step S6, the time measuring unit 12 determines the utterance time t2.

In step S <b> 7, the time measuring unit 12 starts measuring the non-response time t <b> 3.
In step S8, the determination unit 13 determines whether the amplitude of the audio data exceeds the utterance determination threshold value a1. If this determination is YES, the process proceeds to step S10. On the other hand, if the determination is NO, silence continues, and the process moves to step S9.

In step S9, the determination unit 13 determines whether or not the no-response time t3 exceeds a fourth time threshold (4 seconds). If this determination is YES, the process proceeds to step S10. On the other hand, if the determination is NO, the no-response time continues, so the process returns to step S8.
In step S10, the time measuring unit 12 determines the no-response time t3.

In step S11, the determination unit 13 determines that all of the determined silence time t1, speech time t2, and no response time t3 are within a predetermined time threshold range (t1> 30 seconds, 2 seconds>t2> 0.5 seconds, t3). > 4 seconds). If this determination is YES, it is determined that the utterance detected at the utterance time t2 is an inquiry to the system, and the process proceeds to step S12. On the other hand, if the determination is NO, it is determined that the utterance is not an inquiry to the system, and the process returns to step S1.
In step S12, the output unit 14 performs a predetermined response output such as output of the current time as a response to the detected inquiry.

  According to the present embodiment, the utterance detection device 1 determines that the utterance is an inquiry to the device when the silence time t1, the utterance time t2, and the no-response time t3 satisfy predetermined conditions. Therefore, the utterance detection device 1 can easily detect a voice inquiry to the own device without receiving a special operation such as a button press.

  The utterance detection device 1 determines that there is no conversation based on the length of the silence time t1, determines that the utterance is a question based on the length of the utterance time t2, and responds based on the length of the no-response time t3. It is determined that there is no human to do. Therefore, the utterance detection device 1 can easily detect an inquiry to the own device without performing high-load voice analysis.

  Further, since the utterance detection device 1 can adjust the utterance determination threshold value a1 according to the size of the ambient noise, the utterance time t2 can be appropriately measured according to the situation, and the inquiry can be detected more appropriately.

  Moreover, since the utterance detection apparatus 1 outputs a predetermined response in response to detecting the inquiry, the utterance detection apparatus 1 can execute processing triggered by the user's inquiry. Therefore, the user can obtain a response such as reading out the current time, for example, by making a voice inquiry without performing any special operation.

[Second Embodiment]
Hereinafter, a second embodiment of the present invention will be described. In addition, about the structure similar to 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted or simplified.
The utterance detection device 1a of this embodiment confirms that the voice at the utterance time t2 is a human voice.

FIG. 4 is a block diagram showing a functional configuration of the utterance detection device 1a.
The utterance detection device 1 a includes a voice acquisition unit 11, a timing unit 12, a determination unit 13, an output unit 14, and an identification unit 15.

The discriminating unit 15 discriminates whether or not the voice data pattern at the utterance time t2 is based on a human voice, for example, using a predetermined discriminator using machine learning.
For example, the identification unit 15 inputs a frequency pattern obtained by Fourier transform of voice data to the classifier, and at least a human voice and other sounds (for example, footsteps, door knocking sounds, and toilet running sounds) , Dog calls, mobile phone ringtones, etc.). At this time, it is not necessary to perform detailed analysis such as whether the voice is a Japanese syllabary (for example, distinguishing between “a” and “ka”).

  Thereby, the determination unit 13 satisfies the conditions of t1 to t3 in the first embodiment, and the speech time is determined when the identification unit 15 identifies that the voice data pattern at the speech time t2 is based on a human voice. It is determined that the voice at t2 is an utterance for asking the utterance detection device 1a.

  According to the present embodiment, the utterance detection device 1a identifies the voice data pattern of the utterance and determines that it is a human voice. Since the utterance detection device 1a discriminates between human voices and other voices using a discriminator, the accuracy of question detection can be improved without performing high-load voice analysis.

[Third Embodiment]
Hereinafter, a third embodiment of the present invention will be described. In addition, about the structure similar to 1st Embodiment or 2nd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted or simplified.
The utterance detection device 1b according to the present embodiment analyzes utterance contents and executes processing according to the analysis result.

FIG. 5 is a block diagram showing a functional configuration of the utterance detection device 1b.
The utterance detection device 1 a includes a voice acquisition unit 11, a timer unit 12, a determination unit 13, an output unit 14, an identification unit 15, and a voice analysis unit 16.

When it is determined that the voice at the utterance time t2 is a questioned utterance to the utterance detection device 1b, the voice analysis unit 16 performs a predetermined voice analysis process on the voice data that is subsequently input.
Specifically, for example, after outputting a response such as “and?” By the output unit 14, the voice analysis unit 16 analyzes the voice data newly acquired by the voice acquisition unit 11 in detail, Determine the utterance content.

  According to the present embodiment, the utterance detection device 1b performs an analysis process on the subsequently uttered voice when detecting an inquiry to the own device. Therefore, the user can activate a detailed speech recognition function triggered by a voice inquiry to the speech detection device 1b. As a result, the utterance detection device 1b does not need to always operate a high-load voice analysis function, and thus can reduce the load and save power, particularly in a portable terminal.

[Fourth Embodiment]
The fourth embodiment of the present invention will be described below. In addition, about the structure similar to 1st Embodiment to 3rd Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted or simplified.
The utterance detection device 1c according to the present embodiment analyzes the utterance content and executes a process according to the analysis result.

FIG. 6 is a block diagram showing a functional configuration of the utterance detection device 1c.
The utterance detection device 1 a includes a voice acquisition unit 11, a timing unit 12, a determination unit 13, an output unit 14, an identification unit 15, a voice analysis unit 16, and a storage unit 17.

The storage unit 17 temporarily stores voice data at the utterance time t2.
The voice analysis unit 16 performs a predetermined voice analysis process on the voice data stored in the storage unit 17 when it is determined that the voice at the utterance time t2 is a questioned utterance to the utterance detection device 1c.

  According to the present embodiment, the utterance detection device 1c stores voice data of the utterance time t2, and executes analysis processing on the stored voice data when it is determined as an inquiry to the own device. . Therefore, since the user does not speak anew for voice analysis after making an inquiry, convenience is improved.

  As mentioned above, although embodiment of this invention was described, this invention is not restricted to embodiment mentioned above. Further, the effects described in the present embodiment are merely a list of the most preferable effects resulting from the present invention, and the effects of the present invention are not limited to those described in the present embodiment.

The flow of the utterance detection process is an example, and any method may be used as long as it detects an utterance that satisfies the above-described conditions regarding the silence time t1, the utterance time t2, and the no-response time t3.
The utterance detection device 1 may end the utterance detection process when the silence time t1 continues longer than a predetermined time. Further, when a human voice is identified at the utterance time t2, the time until the utterance detection process is ended may be extended.

The speech detection method by the speech detection device 1 is realized by software. When realized by software, a program constituting the software is installed in the information processing apparatus (speech detection apparatus 1). These programs may be recorded on a removable medium such as a CD-ROM and distributed to the user, or may be distributed by being downloaded to the user's computer via a network.
The present invention can be applied to various electronic devices such as a navigation device or a clock device, and the speech detection program may be distributed by being incorporated in a hardware chip that can be arranged in various electronic devices.

DESCRIPTION OF SYMBOLS 1, 1a, 1b, 1c Utterance detection apparatus 11 Voice acquisition part 12 Timing part 13 Judgment part 14 Output part 15 Identification part 16 Voice analysis part 17 Storage part a1 Utterance judgment threshold t1 Silence time t2 Utterance time t3 No response time

Claims (9)

An audio acquisition unit that continuously acquires audio data;
The first time when the volume of the acquired voice is less than the utterance determination threshold, the second time when the volume of the voice exceeds the utterance determination threshold after the lapse of the first time, and the second time A time measuring unit that measures a third time when the volume of the voice is less than the utterance determination threshold after elapse of time,
A determination unit that determines that the voice at the second time is an utterance of an inquiry to the device when the first time, the second time, and the third time satisfy a predetermined condition; An utterance detection device comprising:   The predetermined condition is that the first time is greater than or equal to a first time threshold, the second time is within a range from a second time threshold to a third time threshold, and the third The utterance detection device according to claim 1, wherein the time is a time equal to or greater than a fourth time threshold. A discriminating unit for discriminating whether or not the pattern of the voice data in the second time is based on a human voice by a predetermined discriminator;
When the determination unit satisfies the predetermined condition and the identification unit identifies that the pattern of the voice data at the second time is due to a human voice, the voice at the second time is The utterance detection apparatus according to claim 1, wherein the utterance detection apparatus determines that the utterance is an inquiry to the own apparatus.   4. The utterance detection according to claim 1, wherein the time measuring unit adjusts the utterance determination threshold value based on a statistical value related to a loudness level of the first time and the third time. 5. apparatus.   The utterance according to any one of claims 1 to 4, further comprising: an output unit that outputs a predetermined calculation processing result as a voice when it is determined that the voice at the second time is a questioned utterance to the own device. Detection device.   A speech analysis unit that executes a predetermined speech analysis process on speech data that is subsequently acquired when it is determined that the speech at the second time is an inquiry utterance to the device. The utterance detection device according to claim 5. A storage unit for storing audio data at the second time;
A voice analysis unit that performs a predetermined voice analysis process on the voice data stored in the storage unit when it is determined that the voice at the second time is an utterance of an inquiry to the device. The utterance detection device according to any one of claims 1 to 5. A voice acquisition step for continuously acquiring voice data;
The first time when the volume of the acquired voice is less than the utterance determination threshold, the second time when the volume of the voice exceeds the utterance determination threshold after the lapse of the first time, and the second time A time measuring step of measuring a third time when the volume of the voice is less than the utterance determination threshold after elapse of
A determination step of determining, when the first time, the second time, and the third time satisfy a predetermined condition, that the voice at the second time is an inquiry utterance to the device; A speech detection method executed by a computer. A voice acquisition step for continuously acquiring voice data;
The first time when the volume of the acquired voice is less than the utterance determination threshold, the second time when the volume of the voice exceeds the utterance determination threshold after the lapse of the first time, and the second time A time measuring step of measuring a third time when the volume of the voice is less than the utterance determination threshold after elapse of
A determination step of determining, when the first time, the second time, and the third time satisfy a predetermined condition, that the voice at the second time is an inquiry utterance to the device; An utterance detection program for causing a computer to execute.

Images