JP7296214B2 - speech recognition system - Google Patents

Description

The present invention relates to a speech recognition system with high recognition accuracy for utterances of speakers.

2. Description of the Related Art A speech recognition device that converts voice data input from a microphone into text is widely used. Patent Document 1 discloses a speech recognition apparatus that learns utterance characteristics for each speaker to improve the accuracy of speech recognition.

In addition, learning systems using deep learning are built in various cloud services. These learning systems perform data analysis and analysis by self-learning with neural network-based processing devices based on data collected from a large number of users via the Internet.

By self-learning without waiting for instructions from humans, it is possible to efficiently improve the output accuracy of the processing device, and users can use the analysis results of data analysis by cloud services that utilize deep learning. are doing.

Japanese Patent Application Laid-Open No. 2002-2151848

However, since the speech recognition apparatus of Patent Document 1 learns only from speech collected from a microphone, there is a limit to the speech data that can be collected. Further, in the cloud service described above, a large-scale system is constructed in which a large number of arithmetic processing units are arranged in parallel. Therefore, the speech recognition apparatus of Patent Literature 1 has a problem that improvement in learning accuracy is slow, and accuracy in text conversion and speaker identification is slow to improve.

In addition, since the voice input to the microphone is collected under various recording environments, when the voice is analyzed and converted into text by the processing device, even if the same speaker reads the same sentence, Accuracy of speech recognition varied depending on the recording environment.

The object of the present invention is to solve the above-mentioned problems, and by using a learning server that is a cloud service via the Internet, the collected speech is converted into text with high accuracy, and the speaker can be specified with high accuracy. To provide a recognition system.

A speech recognition system according to the present invention for achieving the above object comprises a processing unit for generating a digital audio file composed of digital audio data based on analog audio data including audio uttered by a speaker; and a monitor unit that displays the processing result. The digital voice file is transmitted to a character conversion server connected via the Internet and equipped with a self-learning function, and the digital voice file is converted to text from the character conversion server. a speech recognition system for receiving a text file converted into text files and displaying the text file on the monitor unit, wherein the processing unit digitizes the analog sound data at a predetermined bit rate and removes noise. A speaker identification server connected via the Internet and having a self-learning function for generating the digital voice data by removing noise based on a first threshold for removing noise and a second threshold for removing noise. , the digital voice file and the speaker's user ID information are transmitted, the speaker's identification result for the digital voice file is received, and the text file and the speaker's identification result are combined with the It is characterized by being displayed on the monitor unit .

According to the speech recognition system according to the present invention, analog sound data including voice input from a sound collecting unit such as a microphone, or analog sound including voice recorded in advance as an electronic file via the sound collecting unit Data is digitized at a bit rate that is easy for a character conversion server on the Internet to recognize, and in order to improve the conversion accuracy of the character conversion server, initial settings are made for noise and noise removal thresholds, Perform noise removal processing. By performing the initial setting process of the threshold value, it is possible to convert the digital voice data into text with high accuracy, and to convert the contents of the conversation into text almost in real time.

1 is a system configuration diagram of a speech recognition system; FIG. FIG. 4 is an explanatory diagram when analog sound data is digitized; FIG. 10 is a flowchart of initial setting processing for finely adjusting the threshold; FIG. 10 is a flowchart for generating an audio file from audio data; FIG. 3 is an explanatory diagram showing voice data of a speaker represented by a waveform; 4 is a list of voice data for each speaker; FIG. FIG. 10 is an explanatory diagram showing voice data of another speaker represented by a waveform; FIG. 4 is an explanatory diagram for determining speech data for each speaker; FIG. 4 is an explanatory diagram of a text sentence displayed on a monitor unit;

The present invention will be described in detail based on the illustrated embodiments. FIG. 1 is a system configuration diagram of a speech recognition system 1. For example, sound collectors such as microphones are arranged at the center of speakers Ha, Hb, and Hc.

The speech recognition system 1 includes a sound collector 2 that inputs ambient sounds as voltage values that change continuously with time, and analog sound data S, which are continuous voltage values input from the sound collector 2, as digital data. A process of generating a digital voice file F by performing conversion processing and noise removal processing, transmitting this digital voice file F to a character conversion server and a speaker identification server, and receiving a text file W and a result of identifying the speaker H. It is composed of a unit 3 and a monitor unit 4 for displaying the processing result of the processing unit 3 .

A commercially available PC, smartphone, or tablet terminal may be used for the speech recognition system 1, and an external microphone or the like is used as the sound collector 2, for example. This sound collecting unit 2 is installed in the center of the table or the like between the speakers.

The processing unit 3 comprises a computing unit 3a, a memory unit 3b and a storage unit 3c, and performs digitization processing and various data processing by starting software stored in the storage unit 3c. The processing unit 3 and the sound collecting unit 2 are connected by wire or wirelessly.

The monitor unit 4 is connected to the processing unit 3 and is composed of, for example, a liquid crystal display. Note that the monitor unit 4 may be a monitor of another PC or mobile terminal connected via a network.

The character conversion server 5 is a neural network-based API (Application Programming Interface) existing on the Internet IN, and is connected to the speech recognition system 1 via the Internet IN.

In the character conversion server 5, when the digital voice file F is uploaded from the processing unit 3 of the voice recognition system 1 via the Internet IN, text conversion processing is performed to convert it into a text file W, and the processing unit 3 generates a text file. You can download W. If a long digital voice file F over several minutes is uploaded, it takes time to convert it into text. Therefore, it is preferable to upload the digital voice file F obtained by dividing the digital voice data V into several tens of seconds or less to the character conversion server 5. .

In addition, the character conversion server 5 performs deep learning based on digital voice files F uploaded by a large number of users, and self-corrects the text conversion process. Therefore, as time passes, the conversion accuracy of the text conversion process improves.

At the same time, the speaker identification server 6, which is separate from the character conversion server 5, is a neural network-based API existing on the Internet IN, and is connected to the speech recognition system 1 via the Internet IN.

In this speaker specifying server 6, a voice sample is registered in advance for each speaker H, and when the digital voice file F is uploaded from the speech recognition system 1 to the speaker specifying server 6, it is processed based on the registered speaker data. , it is possible to identify the speaker H of the digital audio file F. For example, when uploading a digital voice file F of a speaker Ha, if the speaker Ha has already been registered in the speaker identification server 6, the voice of the digital voice file F is identified as the speaker Ha. become. In addition, since a large number of speakers H are registered in the speaker identification server 6, by uploading a group of registered user IDs together with the digital voice file F, it is possible to efficiently speak from the user IDs in the group. Person H can be recognized.

This speaker identification server 6 also analyzes data by self-learning using deep learning based on the digital voice files F uploaded by a large number of users, and identifies the speaker as time passes. The accuracy of speaker identification is improved.

For example, three user IDs in a group consisting of a plurality of speakers Ha, Hb, and Hc are converted from analog sound data S containing voices uttered by speakers Ha, Hb, and Hc to a digital voice file. By uploading together with F to the speaker identification server 6, the identification of each speaker Ha, Hb, Hc will be selected from the user IDs of the three people in the group. Therefore, the speaker identification server 6 can speed up the processing speed of speaker identification and improve the accuracy of speaker identification.

As shown in FIG. 1, for example, when speakers Ha, Hb, and Hc start a conference, one sound collector 2 is arranged at the center of the speakers Ha, Hb, and Hc. The processing unit 3 converts the analog sound data S input from , at any time, to generate a digital voice file F, and the digital voice file F is uploaded to the character conversion server 5 and the speaker identification server 6 .

First, in transmitting the most appropriate digital voice file F to the character conversion server 5, the sound collectors 2 are arranged at equal intervals for the speakers Ha, Hb, and Hc. The distance from each speaker H to the sound collector 2 may be equal, and the distance may be appropriately set within a range of 1 to 3 m, for example, according to the layout of the desk or the like.

The analog sound data S, which is a voltage value that changes continuously with time and is input from the sound collector 2 arranged in this way, is converted into a discontinuous value of the bit rate in the following equation (1). digitization by
Bit rate = sampling frequency x number of quantization bits x number of channels (1)

For the analog sound data S, each value is set so that the character conversion server 5 can easily recognize the bit rate, and data processing for noise removal, which will be described later, is performed.

The higher the sampling frequency in equation (1), the higher the sound range that can be recorded. is preferably set to about 4 KHz or 8 KHz to cut the

While the setting of the sampling frequency is the setting of the interval in the time direction, the number of quantization bits, which is the setting of the interval in the amplitude direction, is set. By setting the number of quantization bits to 16 bits, the voltage value can be quantized in the range of 0 to 65535, that is, digitized.

It is preferable to set the number of channels to 1, which is monaural, and the bit rate of formula (1) thus obtained is 64 Kbps when the sampling frequency is 4 KHz and 128 Kbps when the sampling frequency is 8 KHz. It will be converted to the center.

FIG. 2 is an explanatory diagram when the voltage value, which is the analog sound data S input from the sound collector 2, is digitized according to the bit rate of formula (1). Intervals between output values correspond to sampling frequencies, and the vertical axis corresponds to quantized values obtained by quantizing voltage values corresponding to volume.

Using the speech of one of the speakers Ha, Hb, and Hc, the quantization value converted to the set bit rate is further processed to set threshold values P1 and P2 for noise removal. By initializing these thresholds P1 and P2 and performing noise removal processing, the digital voice file F to be uploaded is generated as digital voice data V that can be easily recognized by the character conversion server 5 and the speaker identification server 6. be able to

As a procedure for removing noise, first, several seconds pass while the speaker H does not speak, and the highest quantized value is stored as the threshold value P1. The quantized value, which is the threshold value P1 in the non-utterance state, is a value obtained by quantizing noise, which is sound that cannot be heard by human ears, such as the sound of an air conditioner in a room, the sound of a computer fan, etc., and is within a certain range. A value with little change over time is obtained.

Subsequently, after the speaker H speaks for a predetermined period after the passage of time during which he has not spoken, the maximum quantized value R can be extracted. A threshold P2 is set. This predetermined ratio is preferably about 1.2, for example.

A quantized value below the threshold value P1 corresponds to the noise described above, and a quantized value above the threshold value P2 corresponds to noise. By removing these noises and noises made up of noises, it is possible to generate digital audio data V consisting only of quantized values included between the thresholds P1 and P2 enclosed by the rectangular dotted lines shown in FIG. can.

Then, the quantized value of the noise Z exceeding the threshold value P2 shown in FIG. , a digital voice file F composed of this digital voice data V is transmitted to the character conversion server 5 and the speaker identification server 6 .

FIG. 3 is a flow chart showing the details of the initial setting process for finely adjusting the threshold values P1 and P2 by the processing unit 3 when creating the digital audio file F. As shown in FIG. In this initial setting process, the speaker H reads out a manuscript prepared in advance, compares the text recognized by the character conversion server 5 with the read out manuscript, and finely adjusts the threshold values P1 and P2. By doing so, the character recognition by the character conversion server 5 can be performed with higher accuracy.

In step ST11, first, speaker H reads out sentence Tx0, which is a prepared manuscript, and inputs it to sound collecting unit 2 after a period of time during which speaker H does not speak. A sentence Tx0 containing a high-pitched tone with a high volume in conversation and a low-pitched tone with a low volume "N", for example, a character string such as "Konnichiwa, Sashisu Seso, Shin Shinto Yukiga Fritsumoru" is prepared in advance. Read aloud as a prepared manuscript.

Subsequently, at step ST12, the above-described digitization processing at the set bit rate is performed on the reference analog sound data S0 obtained by reading out the text Tx0 input via the sound collector 2. FIG. Through this digital conversion process, quantized values as shown in FIG. 2 are obtained.

Subsequently, in step ST13, thresholds P1 and P2 are set for the quantized values, and a digital audio file F0 with quantized values between the thresholds P1 and P2 is created. A threshold value P1 for removing noise is set based on the quantized value of the first non-speech time period, and a threshold value P2 for removing noise is set based on the maximum quantized value in the read-out sentence Tx0.

Next, in step ST14, the digital voice file F0 is uploaded to the character conversion server 5 via the Internet IN. After several seconds, the character conversion server 5 receives the text file W1, and in step ST15, compares the text Tx1 of the text file W1 with the text Tx0 read out in advance. If they match, the process proceeds to step ST16, the threshold values P1 and P2 for the character conversion server 5 and the speaker identification server 6 are stored in the storage unit 3c, and the initial setting process is completed.

In step ST15, if the comparison between the sentences Tx0 and Tx1 does not match, the process proceeds to step ST17. In step ST17, the threshold values P1 and P2 in step ST13 are finely adjusted for the processed sound data S1. This is because the recognition rate of the character conversion server 5 changes depending on the recording environment or the like, so that the threshold values P1 and P2 that are easily recognized by the character conversion server 5 are extracted.

In step ST17, as a result of comparing the sentences Tx0 and Tx1, if the first characters do not match, for example, if the sentence Tx0 is recognized as "onnichiha", the quantization value at the beginning of pronunciation is cut more than necessary. Therefore, a process of lowering the threshold value P1 by several percent is performed. Further, when the recognition rate is low for the processed sound data S1 of "sasisseso", the threshold value P2 is further increased by several percent. Based on the threshold values P1 and P2 adjusted in this manner, noise removal processing is performed to create a digital audio file F0.

Then, in step ST14, the digital voice file F0 is sent to the character conversion server 5 again, and in step ST15, the sentence Tx2 converted to text for the digital voice file F0 is compared with the read-out sentence Tx0.

If the sentence Tx0 and the sentence Tx2 do not match again, the process returns to step ST17. In the correction process of step ST17, the matching rates of the compared sentences are stored, and the rate of increase or decrease of the thresholds P1 and P2 is appropriately adjusted until the sentences Tx0 and Txn match.

In this way, the noise based on the threshold values P1 and P2 and the process of removing the noise are repeated in steps ST14, ST15, and ST17, so that the character conversion server 5 recognizes a digital voice composed of the digital voice data V with a high recognition rate. The file F can be created, and the finely adjusted thresholds P1 and P2 are stored in the storage unit 3c, completing the initial setting process.

FIG. 4 is a flow chart of the processing of the processing section 3 when the conference is started by the speakers Ha, Hb, and Hc after the initial setting is completed. When the conference starts, the speakers Ha, Hb, and Hc chronologically utter their voices as shown in FIG. 5, and one analog sound data S is obtained by synthesizing the voices. Note that this analog sound data S may be electronic file analog sound data S that has been recorded in advance via the sound collector 2, and may be recorded directly via the sound collector 2 as shown in FIG. It may be the analog sound data S input to the processing unit 3 through the process.

In step ST21, the input analog sound data S is subjected to digitization and noise removal processing from steps ST11 to ST13 in FIG. By these digitization processing and noise removal processing, for example, for analog sound data S mixed with air conditioner sound, chair moving sound, ambulance siren sound, etc., those noises are removed as noise Digital Audio data V can be generated.

FIG. 5 is an explanatory diagram in which the digital audio data V whose recording is started at time t11 is represented by a simplified waveform for easy understanding. For example, first, speaker Ha says, "I'm going to start a meeting." Next, speaker Hb says, "I understand." This is the waveform of the voice when Ha said, "Now let's move on to the agenda."

Subsequently, the process proceeds to step ST22 in FIG. 4, and the silent time m, which is a silent state between utterances of the digital voice data V, is measured. For example, when the threshold value of the silent time m is set to 1 second, and the silent time m1 of 1 second or longer occurs, the process proceeds to step ST23. If there is a silent time m0 of one second or less in step ST22, the process of step ST22 is repeated.

At step ST23, it is determined whether or not there are a plurality of speakers H in the immediately preceding digital voice data V separated by the silent time m1. This process of identifying the speaker H utilizes the fact that the center frequency differs for each speaker H in the digital voice data V sampled at predetermined intervals. From the displacement of the center frequency, it is possible to determine the number of speakers H of the immediately preceding voice data separated by the silent time m1.

After determining the number of speakers H, the process proceeds to step ST24. If the number of speakers H is plural, the process proceeds to step ST25. If the number of speakers H is singular, step ST25 is omitted and the process proceeds to step ST26. .

In the digital voice data V shown in FIG. 5, since the immediately preceding digital voice data V divided by times t12, t13, t14, and t15 all have frequency characteristics of each person, step ST25 is omitted from step ST24. Then, the process proceeds to step ST26.

The processing of step ST25 will be described later, and the processing of step ST26 will be described first. Digital voice file F1: t11, digital voice file F2: t12 of "I understand" uttered by speaker Hb, digital voice file F3: t13 of "I understand" uttered by speaker Hc, speaker Ha It is saved as a digital voice file F4:t14 of the uttered "Let's move on to the agenda." It should be noted that the processor 3 cannot specify whose utterances are for these digital audio files F. FIG.

Then, each generated digital voice file F is transmitted to the character conversion server 5 and the speaker identification server 6 . After the transmission, the process returns to step ST22 and repeats the processing of steps ST22 to ST26.

The initial setting process of the thresholds P1 and P2 shown in the flow chart of FIG. 3 can also be set using the analog sound data S of the speaker Ha's "The meeting will start now." Threshold values P1 and P2 can be set for the analog sound data S from the point of silence when recording is started to time t12.

In addition, in order to prevent the character conversion server 5 from erroneously converting "I will start the meeting." A digital audio file F1:t11 corresponding to the divided digital audio data V is created by dividing the preceding data as a starting point.

FIG. 5 also shows that when the speakers Ha, Hb, and Hc are having a conversation as described above, after the first speaker Ha finishes the conversation, after the silent time m1 occurs, the next speaker Hb's voice FIG. 7 shows a simplified waveform of the digital voice data V in the case where the silent time m0 during the conversation of the speakers Ha, Hb, and Hc is equal to or less than the threshold. It is an explanatory diagram of

When the first silent time m1 occurs after speaker Ha's utterance "Now let's move on to the agenda" shown in FIG. On the other hand, the center frequency is measured by sampling at a predetermined time. Then, if there are a plurality of center frequencies in step ST24, that is, if there are a plurality of speakers H, the process proceeds to step ST25.

At step ST25, a digital voice file F is generated for each speaker H that has been identified. FIG. 8 is an explanatory diagram for determining the speaker H from the displacement of the center frequency for the immediately preceding digital voice data V' separated by the occurrence of the silent time m1 at time t25. By determining the center frequency of this digital audio data V', it can be divided into digital audio data VH1 starting at time t21, digital audio data VH2 starting at time t22, and digital audio data VH3 starting at time t23. .

Note that even if two speakers H are redundantly uttered and recorded in a part of the digital voice data V′, by increasing the sampling frequency, the speakers occupying each sampling time H can be specified, and it is possible to partition the digital voice data V' that are uttered in duplicate into individual digital voice data V. FIG.

Furthermore, two digital voice files F1:t21 and F4:t24 can be generated from the digital voice data VH1 consisting of two utterances, since there is a silent time m1 between the utterances.

By performing the determination process described above, digital audio files F1:t21 to F4:t24 similar to the digital audio files F1:t11 to F4:t14 shown in FIG. 6 can be generated. Although the processing unit 3 can determine that the speakers are different for these digital audio files F1:t21 to F4:t24, it cannot specify whose utterances are.

Further, it is also possible to generate only one digital audio file F1:t21 without performing the process of generating two digital audio files F from the digital audio data VH1. In this case, in order to clarify the chronological sequence between the utterances in the second half of the digital voice file F1: t21 and the utterances of the digital voice files F2: t22 and F3: t23, information of times t21 to t24 is stored in each voice file. There is a need to. That is, by storing the times t21 and t24 in the digital voice file F1:t21, each utterance can be displayed in chronological order from time t21 to t24 on the monitor section 4, which will be described later.

5 and the digital audio data V shown in FIG. 7, the digital audio data V shown in FIG. The digital audio files F1:t11 to F4:t14 in FIG. 6 are generated in order from the top, and each time they are generated, the process proceeds to step ST26. On the other hand, in the digital audio data V' of FIG. 7, the digital audio files F1:t21 to F4:t24 are generated almost simultaneously, and the process proceeds to step ST26.

In step ST26, when the generated digital voice file F is sent to the character conversion server 5, the digital voice files F1: t11 to F4: t14 and the digital voice files F1: t21 to F4: t24 are converted into text files. W1: t11 to W4: t14 and text files W1: t21 to W4: t24, and the speech recognition system 1 receives these files.

Further, when transmitting the generated digital voice file F to the speaker identification server 6, in addition to the digital voice files F1: t11 to F4: t14 and the digital voice files F1: t21 to F4: t24, The user IDs of the speakers Ha to Hc are also transmitted. The speaker identification server 6 selects from among the user IDs sent together with the digital voice files F1:t11 to F4:t14 and the digital voice files F1:t21 to F4:t24 sent from the processing unit 3 Each speaker H is identified, and the processor 3 receives the identified speakers Ha to Hc corresponding to the digital voice file F. FIG.

Then, the processing unit 3 associates the text file W and the identified speaker H with the digital voice file F, and displays them on the monitor unit 4 in chronological order. In other words, the digital voice file F1:t11 of "I will start the meeting now.", for which speaker H could not be specified, is replaced with the text file W1:t11 of "I will start the meeting now." , are displayed as shown in FIG.

The digital voice file F is stored as a serial number corresponding to time t at the end of the file name, and the text file W and speaker H are displayed in chronological order of time t as shown in FIG. In FIG. 9, in order to facilitate identification of speaker H, speaker Ha is displayed on the left side, and speakers Hb and Hc are displayed on the right side.

Although some time lag occurs in the generation of each file and transmission/reception to the cloud service in this way, the date and time of the latest digital voice file F, the text file W, and the speaker H are displayed almost in real time from the bottom of the screen of the monitor unit 4. will be displayed in order.

Although the display order is determined based on the file name of the digital audio file F, by storing the times t11 to t14 in the file header etc. can be displayed in chronological order.

Also, instead of the displayed speech date and time, the processing date and time of display processing may be displayed on the screen. In this case, without storing the above-mentioned utterance date and time, the processing unit 3 transmits the digital voice files F in the order in which they were generated to the cloud service, and on the condition that they are received, the next digital voice file F is sent to the cloud service. You may do so.

As shown in FIG. 9, by displaying both the text file W and the identification result of the speaker H as the speaker conversation format, it is possible to easily confirm who said what kind of statement at a later date. be. Also, by enabling the screen shown in FIG. 9 to be viewed on a PC or a portable terminal connected to another terminal device, it is possible to visually confirm the content of the conference from another location in almost real time.

In particular, when listening to the conference by sound at another place, it is difficult to identify the speaker H and to grasp the entire contents of the conference. It is easy to understand the content of the meeting because it can be confirmed by text.

Furthermore, it is possible to easily grasp the contents of the conference when confirming the contents of the conference in a place where it is difficult to output sound, or when confirming the contents of the conference by a hearing-impaired person. By scrolling the screen, you can easily check past remarks.

In addition, although the analog sound data S collected by the sound collecting unit 2 installed in the center of the speakers Ha to Hc such as the table of the speech recognition system 1 is used in the description, the voice recorded at another place etc. is also included. A file of analog sound data S is stored in the storage unit 3c via a network, a storage medium, or the like, or read directly from the storage medium, etc., and the processing of the flowcharts shown in FIGS. 3 and 4 is performed by the calculation unit 3a. can be

As described above, the speech recognition system 1 includes the character conversion server 5 and the speaker identification server 6, which are cloud services for performing data analysis and analysis by self-learning based on data collected from a large number of users via the Internet IN. By using it, accurate character conversion and speaker identification can be performed without providing a character conversion function and a speaker identification function.

In addition, the processing unit 3 processes analog sound data S including sound input from the sound collecting unit 2 such as a microphone, or analog sound data S including sound recorded in advance as an electronic file via the sound collecting unit 2. S is digitized in accordance with a bit rate that is easy for the character conversion server 5 to recognize, and in order to improve the conversion accuracy of the character conversion server 5, noise and threshold values P1 and P2 for removing noise are initialized, Perform noise removal processing. By performing the initial setting processing of these thresholds P1 and P2, the digital voice data V can be converted into text with high accuracy, and the contents of the conversation can be converted into text almost in real time. Furthermore, since the recorded voice file can also be converted into text after the fact, it is possible to make use of it for quick comprehension of the content of the meeting.

1 voice recognition system 2 sound collector 3 processor 4 monitor 5 character conversion server 6 speaker identification server IN Internet

Claims (7)

Consists of a processing unit that generates a digital audio file composed of digital audio data based on analog sound data including the voice uttered by a speaker, and a monitor unit that displays the processing result of the processing unit,
The digital voice file is transmitted to a character conversion server connected via the Internet and equipped with a self-learning function, the text file obtained by converting the digital voice file into text is received from the character conversion server, and the text file is converted to text. A voice recognition system displayed on the monitor unit,
The processing unit digitizes the analog sound data at a predetermined bit rate, and removes noise based on a first threshold for noise removal and a second threshold for noise removal. generating the digital audio data;
Sending the digital voice file and user ID information of the speaker to a speaker identification server connected via the Internet and equipped with a self-learning function, and receiving the speaker identification result for the digital voice file. and displaying the text file and the result of identifying the speaker together on the monitor unit . 2. The method of claim 1, wherein the first threshold is set based on a quantized value digitized according to the predetermined bit rate when the speaker is not speaking. voice recognition system. 3. The method according to claim 1, wherein the second threshold is set based on a quantized value digitized according to the predetermined bit rate when the speaker speaks. A speech recognition system as described. 4. The speech recognition system according to any one of claims 1 to 3, wherein said predetermined bit rate is a value converted around a frequency band of human speech. The digital voice data is divided when a silent state between utterances of the speaker is longer than a predetermined time, and the digital voice file is generated based on the digital voice data immediately before the division. The speech recognition system according to any one of claims 1 to 4, wherein: when a silent state between utterances of the speaker is longer than a predetermined time, dividing the digital voice data from a point before the starting point of the utterance, and based on the digital voice data immediately before the division, 6. The speech recognition system of claim 5, wherein the speech recognition system generates the digital speech file. 2. The speech recognition system according to claim 1 , wherein said monitor unit associates said sentence file corresponding to said digital speech file with said speaker of said specified result and displays them in chronological order.

Images