JP6546288B2 - Voice noise detection device and voice noise detection method - Google Patents

Description

  The present invention relates to an audio noise detection apparatus and an audio noise detection method.

There are various features of video and audio in a video server system for broadcast service.
For example, in a broadcast station, a video frame rate of about 29.97 fps (drop frame) is generally employed. In addition, a sampling rate of 48 kHz (48000 samples / second) is generally adopted for voice.

  In video server systems, material data is often handled in units of one frame. At 29.97 fps and 48 kHz, the number of audio samples per unit time (one frame) is 1601.6 samples, resulting in fractions. Therefore, the problem is solved by regularly changing in the range of 1601 samples or 1602 samples per frame. For example, a sequence of 1602 samples in the first, third, and fifth frames and 1601 samples in the second and fourth frames is used with a period of 5 frames as a cycle. Then, every five frames, a frame of 1602 samples will be continuous twice.

  Therefore, it is necessary to recognize whether the current frame is 1602 samples or 1601 samples at the sending side and the receiving side of voice data. In order to indicate that, for example in ANSI / SMPTE 272M, it is possible to insert an integer audio frame number between 1 and 5 in the Audio control packet.

  However, verification means such as CRC (Cyclic Redundancy Check) and checksum is always provided for audio data on HD-SDI (High Definition-Serial Digital Interface) signal which is original data and audio data stored in the storage server itself. It is not always the case. Generally, once voice data is received, only real data without a header is stored in the memory. As a result, if the sender or receiver misunderstands the sequence for some reason, or if the amount of input voice data is different from the amount of data according to the above sequence, this synchronization may result, resulting in data loss. sell. This can happen when audio sources are switched or edited in a journalistic or time-shifting system where perfect editing is difficult.

  FIG. 3 is a diagram for explaining voice data transferred to a server or the like. Here, it is assumed that the encoder accumulates 16-bit / sample uncompressed PCM audio data for each frame and transfers it to a server together with encoded video data and the like. The encoder application reads out the PCM audio data written in the shared memory frame by frame. The read amount changes in the above-mentioned sequence. The receiving server secures a sufficient amount of recording area for the voice of each frame, and records the received voice as it is. FIG. 3A shows the case where the amount of stored data transferred from the encoder and stored in the server is smaller than the original amount of audio data.

  For example, assume that the encoder recognizes the current frame as a frame of 1601 samples. However, 1602 samples of PCM audio data are written in the shared memory. As a result, the encoder application transfers only 2102 bytes of the PCM audio data having 3204 bytes, and the video server loses without storing (storing) 2 bytes of audio data.

  FIG. 3B shows the case where the amount of stored data is larger than the amount of audio data. For example, while the encoder recognizes the current frame as a frame of 1602 samples, only 1601 samples of PCM audio data are written in the shared memory. Such a phenomenon may occur, for example, when video and audio are input as separate physical I / Fs. As a result, the encoder application transfers 3204 bytes of data in which a null value (0x00) of 2 bytes is added to the original 3202 bytes of PCM audio data, and the video server transfers 2 bytes of data that was not present in the original. The extra null value (0x00) is recorded (stored). The last two bytes become null values because the memory is initialized for each frame.

  As described above, since the voice data has a variable length, excess and deficiency of data tends to occur at frame boundaries. Excess or deficiency is recorded as audio noise in the material, and if played as it is, it contributes to broadcast failure. However, it is difficult to identify speech noise even with sophisticated analysis means. Also, in real-time devices in video server systems, it is necessary to avoid high processing load such as analysis as much as possible. Data excess is a state in which initialization data is added to valid data. Even if the initialization data is 0x00, 0x00 does not necessarily indicate noise occurrence because silence data is continuous data of 0x00.

  As a prior art document, for example, Patent Document 1 discloses an invention which performs sampling conversion processing by regarding the number of audio samples of one frame based on the output frame synchronization signal as one frame unit.

Patent registration 3825677 gazette JP, 2010-028223, A Patent Registration No. 4122264 JP, 2000-307645, A Patent registration 4812171 gazette

  An object of the present invention is to discriminate noise generated at frame boundaries of audio data with low load.

  An audio noise detection apparatus according to the present invention is an audio noise detection apparatus for detecting noise caused by audio data being divided at a video frame period and a sample of a known value being inserted into the boundary portion of the division. A volume determination unit that determines the volume for each frame based on a boundary portion determination unit that detects a known value in the vicinity, and a sample other than the vicinity of a frame boundary, and the determined volume exceeds a predetermined value and is a known value And a noise determination unit that counts noise in a frame in which is detected and detects noise based on the count value.

  The known value is 0, NULL, or a value of initialization data, and the boundary portion determination unit determines the frame boundary portion at an interval of a noninteger value obtained by dividing the audio sample rate by the video frame, and determines the volume. The unit is characterized in that the volume is determined to be equal to or less than a predetermined value when the samples other than those near the frame boundary are equal to the known value.

  Furthermore, the audio noise detection method according to the present invention is an audio noise detection method for detecting noise from frame boundaries of audio data, wherein data in frame boundaries of audio data is checked, and other than frame boundaries of audio data. The data is checked, and if the frame boundary portion of the audio data is 0x00, it is determined whether the volume exceeds a predetermined value, and if the volume exceeds a predetermined value, 0x00 is counted and the count value is predetermined In the case of the value or more, it is characterized that it is determined that there is noise.

  According to the present invention, noise generated at a frame boundary portion of audio data can be determined at low load.

FIG. 1 is a block diagram of an audio noise detection apparatus according to an embodiment. The flowchart for demonstrating the operation | movement of the audio | voice noise detection apparatus which concerns on one Example. The figure explaining the audio | voice data transferred to a server etc. FIG.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In FIG. 1, the audio noise detection device 1 includes a boundary portion determination unit 11, a sound volume determination unit 12, a noise determination unit 14, and a noise reduction unit 15. Here, it is assumed that the initialization data of the null value causing the noise appears only one sample at the frame boundary and is not continuous. The audio noise detection device 1 may detect audio noise by software processing using a CPU (Central Processing Unit) or the like.

  In the shared memory 10, a stream of audio data is written from an interface (not shown) such as an SDI or storage, or another process (for example, a decoding process) executed by the CPU, and the audio data is sent to the audio noise detection device 1. Provide readability. Other than shared memory, inter-process communication methods of various abstraction levels such as memory mapped I / O, memory mapped file, and related APIs (Application Programming Interface) can be used. The APIs that can be used with these are limited to the open and close of the stream, the initialization instruction, the flag indicating whether the read is ready, the flag indicating whether the read has reached the end of the data, or Even such an API is partially unavailable.

  The boundary portion determination unit 11 performs data check on audio data read out from the shared memory only at sample positions (the beginning and the end of about 1600 samples) serving as frame boundary portions, as well as the read audio data. It passes to the volume determination unit 12. A frame boundary is set at a 1601.2 sample interval, and one or several samples continuous around it are extracted for checking. In other words, if the boundary of the 1601.2 sample interval comes between two samples, the two samples may be targeted, and if the timing of just one sample matches, the one sample may be targeted. In consideration of multiple occurrences of frame boundary deviation and initialization data mixing, three or more samples may always be extracted. When voice data is input in a burst manner in frame units, the frame boundary is clear, and it is sufficient to check only the last one sample of the burst or three samples including before and after it. As a result of the check, if the target data includes a null value sample or other initialization data, the noise determination unit 14 is notified of this.

  The volume determination unit 12 has an intermediate value of at least one sample, for example, about 1600 samples, in addition to the frame boundary portion of the audio data received from the boundary portion determination unit 11 for the purpose of preventing false detection of 0x00 as noise in the silent section. Check if the data is above the specified volume. For example, if the data is a null value, it is determined that there is no sound (volume is below a predetermined level) in the vicinity of the frame, and other than that it is determined to be sound (a volume exceeds a predetermined level) and the result is output to the noise determination unit 14 . The probability that a sample in a spoken frame is 0x00 is negligibly small. Also, the volume determination unit 12 passes the audio data received from the boundary portion determination unit 11 to the noise reduction unit 15 as it is.

  The noise determination unit 14 has a counter, and increments the counter each time it is notified from the boundary portion determination unit 11 that initialization data such as a null value has been detected. However, when the determination result of the volume determining unit 12 indicates that the volume is equal to or less than a predetermined level, it is determined as a silent section and is not incremented. Therefore, even if a null value continues, it is not considered as noise generation. Further, if not silent and if the count number is equal to or more than a predetermined value with respect to the number of frames, it is determined that noise is generated due to processing abnormality, and a noise detection result is output. The noise detection result includes, if necessary, information indicating the timing of the sample, such as the null value detected by the boundary portion determination unit 11. On the other hand, the noise determination unit 14 unconditionally decrements the counter (but does not become a negative number) in a cycle of a predetermined number of frames of 2 or more. This makes it possible to judge by simple threshold processing of the count number.

  When receiving the noise detection result indicating noise generation, the noise reduction unit 15 replaces the sample of the initialization data such as the null value with the estimated value based on the surrounding samples. The estimated value s2 must be calculated with a processing delay equal to or less than a predetermined value for real-time processing, and can be obtained, for example, as s2 = 2 × s1−s0. Here, s1 is a sample value immediately before s2, and s0 is a sample value immediately before s2. If there is no requirement for a small delay, estimated values obtained by inputting a large number of samples before and after the sample to be replaced into an FIR filter or the like may be used.

FIG. 2 is a flow chart for explaining the operation of the speech noise detection apparatus according to one embodiment of the present invention.
The voice noise detection device 1 reads voice data (S201).
In the process of S202, the data check is performed on only the sample positions (the beginning and the end of the approximately 1600 samples) serving as the frame boundary portion.

In the process of S203, the data is checked with reference to the middle of about 1,600 samples by about one sample other than the frame boundary.
In the process of S204, it is determined whether the audio data is 0x00. If the audio data is 0x00 (YES), the process proceeds to the process of S205. If the audio data is not 0x00 (NO), the process of S210 is performed. move on.

In the process of S205, in order to prevent false detection of 0x00 as noise in a silent section, it is determined whether or not intermediate data of about 1600 samples has a value exceeding a predetermined volume, and the volume is less than a predetermined value ( If the sound volume exceeds the predetermined value (YES), the process proceeds to step S207.
In the process of S206, a silent interval is determined, and even if 0x00 and so on continue, the process proceeds to the process of S210 without being regarded as noise generation.

In the process of S207, the process is counted as if it is 0x00 or other initialization data, and the process proceeds to S208.
In the process of S208, it is determined whether or not the count number is equal to or more than a predetermined value. If the count number is equal to or more than the predetermined value (YES), the process proceeds to S209, and if the count number is less than the predetermined value (NO) The process proceeds to the process of S210.

In the process of S209, the noise detection result determined to have noise due to the process or more is output at the frame boundary portion.
In the process of S210, the process proceeds to the next process.

  When the voice noise detection device according to an embodiment of the present invention is applied to a recording device, detection is performed on voice data to be recorded, alarm notification to the monitoring device, and a dropout sample are corrected depending on cases. Issue (output) a correction task instruction to the device that is recording or managing the system.

  When the audio noise detection apparatus according to an embodiment of the present invention is applied to an apparatus using materials such as a reproduction apparatus and an editing apparatus, detection is performed on audio data used for reproduction / editing etc. Alarm notification, and, if necessary, correct (drop out) a corrected sample and use or issue (output) a corrected task instruction to a device which manages the system.

  The voice noise detection apparatus and the voice noise detection method according to the embodiment of the present invention are stateless verifications that do not have an internal state corresponding to a sequence consisting of frames of 1601 and 1602 samples. And there is no risk of malfunction. In addition, noise generated at frame boundaries of audio data can be determined with extremely low load.

  As mentioned above, although one embodiment of the present invention was described in detail, the present invention is not limited to the embodiment mentioned above, and can be variously changed and carried out in the range which does not deviate from the meaning of the present invention.

  A system based on SMPTE 272M and 299M that embeds audio data in an SDI signal, and that conforms to SMPTE 2022-6 that encapsulates the SDI signal in an IP packet, etc. And suitable for devices and the like.

  10: voice noise detection device, 11: boundary portion determination unit, 12: volume determination unit, 14: noise determination unit, 15: noise reduction unit.

Claims (3)

An audio noise detection device for detecting noise generated by audio data being divided at a video frame period and insertion of a sample of a known value at a boundary portion of the division,
A boundary portion determination unit that detects the known value near the frame boundary;
A volume determination unit that determines a volume for each frame based on samples other than the vicinity of the frame boundary;
An audio noise detection device comprising: a noise determination unit that counts in a frame in which the determined volume exceeds a predetermined value and the known value is detected, and determines the presence or absence of noise based on the count value. The known value is 0, NULL, or a value of initialization data,
The boundary portion determination unit determines the frame boundary portion at a noninteger value interval obtained by dividing the audio sample rate by the video frame,
The volume determining unit is configured to determine that the volume is equal to or less than the predetermined value when a sample other than the vicinity of the frame boundary has the same value as the known value.
The noise reduction device according to claim 1, wherein the voice noise detection device replaces the known value with estimated values based on a plurality of samples around the known value in real time when the noise determination unit determines the occurrence of noise. The speech noise detection apparatus according to claim 1, further comprising: A speech noise detection method for detecting noise from frame boundaries of speech data, comprising:
Check the data of the frame boundary portion of the audio data, check the data other than the frame boundary portion of the audio data, and if the frame boundary portion of the audio data is 0x00, determine whether the volume exceeds a predetermined value, A voice noise detection method comprising: counting 0x00 when the value exceeds a predetermined value, and determining that there is noise when the count value is equal to or more than the predetermined value.

Images