JP4531350B2 - Voice input device and voice recognition processing system - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a voice input device that converts an input voice signal into voice data to be subjected to voice recognition processing.
[0002]
[Prior art]
A voice recognition device for recognizing the contents of voice collected by a microphone is known and applied to an input device of an in-vehicle navigation device. In such a speech recognition device, the content of the speech naturally uttered by the user has not yet been fully recognized, and various devices for increasing the recognition rate have been made. For example, a speech recognition device is known in which the gain of an amplifier that amplifies input speech is set in accordance with the dynamic range of the input speech to adjust the amplitude of the speech to be recognized (see, for example, Patent Document 1). .) In this speech recognition apparatus, the gain of the amplifier is set high for a user with a low voice, and conversely, the gain of the amplifier is set low for a user with a loud voice. The dynamic range can always be maintained at the optimum level, and the recognition rate can be increased.
[0003]
[Patent Document 1]
JP 61-180296 A (2nd page, FIG. 1)
[0004]
[Problems to be solved by the invention]
By the way, in the speech recognition apparatus disclosed in Patent Document 1 described above, the gain of the amplifier is set based on the dynamic range of the input speech, and the optimum dynamic range is set for the subsequent input speech. There is a problem that the recognition rate cannot be increased from the first input until the amplifier gain setting is completed. In addition, it cannot be applied when the dynamic range of the input speech itself changes, such as when the same user or multiple users utter loud and small voices alternately, and the recognition rate cannot be increased. There was a problem.
[0005]
In general, it is said that the dynamic range of human voice is about 60 dB from whispering to yelling. Moreover, considering that the loudness of the voice varies from person to person, the dynamic range of the entire voice is considered to be further increased. When such input voice is converted into voice data using a 16-bit quantization analog-digital converter that is generally used, it corresponds to data in the range of 15 bits to 5 bits.
[0006]
On the other hand, the dynamic range of input speech that can be recognized by the speech recognition processing is currently limited to about 40 dB, and when a 16-bit quantization analog-to-digital converter is used, the dynamic range is changed from 15 bits to 9 bits. Equivalent to. That is, if a voice corresponding to a small voice is input while a gain corresponding to a loud voice is set, this content cannot be recognized.
[0007]
The present invention has been created in view of such a point, and an object thereof is to provide a voice input device capable of increasing the recognition rate for voice having a wide dynamic range.
[0008]
[Means for Solving the Problems]
  In order to solve the above-described problem, a voice input device according to the present invention is a voice input device that is provided in a preceding stage of a voice recognition device and generates voice data corresponding to an input voice signal,By attenuating or amplifying with a predetermined gainA signal generating means for generating a plurality of audio signals having different amplitudes, a plurality of analog-digital converting means for converting each of the plurality of audio signals generated by the signal generating means into digital data, and a plurality of analog-digital conversions Data synthesizing means for synthesizing a plurality of digital data output from the means;And level detection means for detecting the level of each of the plurality of audio signals generated by the signal generation means, and the data synthesis means each bit of the plurality of digital data output from the plurality of analog-digital conversion means The position is synthesized by shifting the number of bits according to the level ratio of each of the plurality of audio signals detected by the level detection means.ing. As a result, it is possible to perform coding processing without a partial loss of a waveform even for an audio signal with a wide dynamic range in which the number of quantization bits of one analog-digital conversion means is insufficient, In addition to ensuring a wide dynamic range, it is possible to increase the recognition rate of the speech recognition process by generating data including the entire speech waveform and inputting it to the speech recognition apparatus.Also, by attenuating or amplifying the input audio signal with a predetermined gain, a plurality of audio signals having different amplitudes (gains) can be easily generated for one input audio signal. In addition, level detection means for detecting the level of each of the plurality of audio signals generated by the signal generation means is provided, and the data synthesizing means is in accordance with the ratio of the levels of the plurality of audio signals detected by the level detection means. Since the number of bits shifted during the synthesis process is determined, it is possible to synthesize data that takes into account element constants and manufacturing variations, and generates data with low distortion corresponding to the input audio signal. can do.
[0009]
It is desirable to further include a microphone that collects sound and outputs an input sound signal. As a result, it is possible to directly convert various sounds collected by the microphone with a wide dynamic range into digital data for speech recognition.
[0014]
The number of analog-digital conversion means described above is two, and it is desirable to use a stereo analog-digital converter. Accordingly, the cost of components can be reduced by using a set of two analog-digital converters that are generally commercially available for stereo use.
[0015]
Further, the data synthesizing means described above accumulates digital data input during a predetermined period, and stores data of a predetermined number of bits including an audio signal having the largest amplitude within this period from among the accumulated digital data. It is desirable to perform a process of cutting out and outputting. As a result, it is possible to generate data of a predetermined number of bits including the peak of the speech waveform necessary for speech recognition. In addition, by generating data with a predetermined number of bits, it is possible to use a conventional voice recognition device that performs voice recognition processing on data with a predetermined number of bits, thereby suppressing an increase in the cost of the entire voice recognition processing system. It becomes possible.
[0016]
Moreover, it is desirable that the predetermined period described above is a voice input section until the input voice signal is interrupted. This makes it possible to cut out a predetermined number of bits of data holding peak information included in the waveform of a series of voices that are subject to voice recognition, thereby increasing the recognition rate.
[0017]
  Further, the predetermined period is a divided period shorter than the voice input period until the input voice signal is interrupted, and the amplitude of the voice signal corresponding to the subsequent divided period is the amplitude of the voice signal corresponding to the previous divided period. If it is larger thanChange the cutout position of the data of the predetermined number of bits to cut out to a position corresponding to this large amplitude,It is desirable to repeat the process of cutting out and outputting from each stored digital data from the previous divided period. As a result, the delay time of the voice recognition process can be reduced.
[0018]
  The speech recognition processing system of the present invention includes the speech input device described above and a speech recognition device that performs speech recognition processing on data output from the speech input device.For digital data input during a predetermined period, the position where data of a predetermined number of bits is cut out is fixed,The voice recognition apparatus is configured to generate digital data output from a plurality of analog-digital conversion means when data output from the data synthesis means is generated using digital data output from one analog-digital conversion means. A plurality of acoustic dictionaries used for the speech recognition process are used differently depending on whether they are used. In particular, the above-described plurality of acoustic dictionaries include a distortion learning acoustic dictionary that takes into account distortion generated when synthesizing digital data output from one analog-to-digital converter, and a normal acoustic that does not consider this distortion. It is desirable to include a dictionary. As a result, it is possible to perform speech recognition processing in consideration of the distortion that occurs with the synthesis of data, and the recognition rate can be increased.
[0019]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, a voice input device according to an embodiment to which the present invention is applied will be described in detail with reference to the drawings.
[First Embodiment]
FIG. 1 is a diagram illustrating the configuration of the voice input device according to the first embodiment. A voice input device 100 according to the present embodiment shown in FIG. 1 is provided in the preceding stage of a voice recognition device and is used to generate voice data corresponding to an input voice signal. A microphone 10, an amplifier 12, and an analog-digital conversion. (A / D) 14 and 18, an attenuator 16, and a waveform estimator 20. In addition, the voice recognition processing system includes the voice input device 100 and the voice recognition device 200 connected to the subsequent stage.
[0020]
The microphone 10 collects the voice of a user who is a voice recognition target and outputs an input voice signal corresponding to this voice. The amplifier 12 amplifies the input audio signal with a predetermined gain so that the amplitude level can be processed by the analog-digital converters 14 and 18. One analog-digital converter 14 converts the amplified audio signal output from the amplifier 12 into digital data having a predetermined number of bits. For example, the audio signal is converted into audio data (intermediate data) having a code bit of 1 and a data bit of 15 in total, ie 16 bits.
[0021]
The attenuator 16 attenuates the audio signal output from the amplifier 12, and outputs the attenuated audio signal. For example, the attenuation gain is (1/2)¹⁵It is set to double. The other analog-digital converter 18 converts the attenuated audio signal output from the attenuator 16 into digital data having a predetermined number of bits. For example, like the one analog-to-digital converter 14, the audio signal is converted into 16-bit audio data (intermediate data) having a code bit of 1 and a data bit of 15. Note that the cost of components can be reduced by using a set of two analog-digital converters 14 and 18 that are generally commercially available for stereo use.
[0022]
The waveform estimator 20 synthesizes 16-bit intermediate data output from each of the two analog-to-digital converters 14 and 18 to generate 31-bit combined data with 1 sign bit and 30 data bits. To do. For example, the waveform estimation unit 20 uses this output data when the output data of the analog-to-digital converter 14 is not saturated, and when this output data is saturated, the waveform shape of this saturated portion is changed to another analog waveform. A synthesis process of intermediate data is performed by estimating based on the unsaturated output data of the digital converter 18.
[0023]
FIG. 2 is a diagram illustrating a detailed configuration of the waveform estimation unit 20. As shown in FIG. 2, the waveform estimation unit 20 includes a double precision data generation unit 22, a speech segment end determination unit 24, a valid data position monitoring unit 26, and a recognition processing data generation unit 28.
As described above, the audio signal input to the other analog-digital converter 18 has a signal level of (1/2) with respect to the audio signal input to one analog-digital converter 14.¹⁵Since these sound signals are attenuated by a factor of 2, when the sound signals are converted into 31-bit digital data, the positions at which the waveform information of each sound signal appears are shifted by 15 bits. In practice, the analog-to-digital converters 14 and 18 can only convert the input audio signal into 15-bit data.
[0024]
For this reason, when an audio signal having a large signal level is input, the one analog-digital converter 14 exceeds the allowable input voltage and is converted into audio data in a state where the peak portion of the audio waveform is saturated. The At this time, in the other analog-to-digital converter 18, since the audio signal having a large signal level is input in a attenuated state, the peak portion of the audio waveform is normally converted into audio data.
[0025]
In addition, when an audio signal having a small signal level is input, the audio signal within the allowable input voltage range is input to one of the analog-to-digital converters 14, so that the peak portion of the audio waveform is normal. Converted to data.
When the intermediate data output from one of the analog-digital converters 14 is not saturated, the double precision data generation unit 22 generates 30 bits of data using the intermediate data as it is, When the intermediate data output from the digital converter 14 is saturated, the intermediate data output from the other analog-digital converter 18 is 2¹⁵The data bits are multiplied to generate 30 data bits, and the code bit 1 and the data bit 30 generate a total of 31 bits of double precision data (synthesized data). As the sign bit included in the combined data, the sign bit of the intermediate data output from the analog-digital converters 14 and 18 is used as it is.
[0026]
As another operation example of the double-precision data generation unit 22, intermediate data output from the other analog-digital converter 18 is added to the upper 15 bits of the data bits included in the 31-bit audio data. And the data bits in the intermediate data output from one of the analog-to-digital converters 14 are applied to the lower 15 bits of the data bits included in the 31-bit audio data (this intermediate When the data is saturated, the lower 15 bits are set to “0”), and combined data of 31 bits in total with the sign bit 1 and the data bit 30 may be generated.
[0027]
The voice segment end determination unit 24 determines the end timing of the voice segment by monitoring the synthesized data generated by the double precision data generation unit 22. For example, after the input of the voice signal is started, when the value of the synthesized data becomes “0” or smaller than a predetermined value, it is determined that the voice input as a recognition target is completed.
[0028]
The valid data position monitoring unit 26 examines each value of 30 bits included in the composite data, detects the most significant bit position where the value is “1”, and extracts the bit position as the valid data position. . This valid data position is updated if the valid data position corresponding to the next input composite data is on the higher bit side, and is discarded in other cases. In this way, the effective data position corresponding to the audio data having the highest signal level is held.
[0029]
The recognition processing data generation unit 28 accumulates the synthesized data output from the double-precision data generation unit 22 until the voice segment end determination unit 24 determines the end timing of the voice segment. Further, the recognition processing data generation unit 28 determines the extraction position of the lower 15 bits including the effective data position detected by the effective data position monitoring unit 26 after the end of the accumulation period, reads the synthesized data in the accumulation order, and 15-bit data corresponding to the extraction position is extracted, and further 16-bit recognition processing data is generated by adding a sign bit. The recognition processing data generated in this way is input to the speech recognition device 200 connected to the subsequent stage of the speech input device 100.
[0030]
The amplifier 12 and the attenuator 16 described above correspond to signal generation means and gain change means, the analog-digital converters 14 and 18 correspond to analog-digital conversion means, and the waveform estimation unit 20 corresponds to data synthesis means.
As described above, in the audio input device 100 according to the present embodiment, two analog-digital converters can be used for an input audio signal having a wide dynamic range in which the number of quantization bits of one analog-digital converter is insufficient. 14 and 18 makes it possible to perform an encoding process with no partial omission of the waveform, secure a wide dynamic range, and generate data including the entire speech waveform and input it to the speech recognition apparatus 200. By doing so, it becomes possible to increase the recognition rate of the voice recognition processing.
[0031]
Further, by using each of the amplifier 12 and the attenuator 16 alone or in combination, it is possible to generate two audio signals having different amplitudes (gains) for one input audio signal.
Further, the waveform estimation unit 20 synthesizes two intermediate data by shifting the number of bits corresponding to the gain of the attenuator 16, thereby generating synthesized data having a large number of bits including the entire waveform of the original input speech signal. It can be easily generated.
[0032]
Also, a conventional speech recognition apparatus that performs speech recognition processing on data having a predetermined number of bits by cutting out and outputting recognition processing data having a predetermined number of bits from synthesized data having a large bit length in the waveform estimation unit 20 Since 200 can be used, it is possible to suppress an increase in cost of the entire speech recognition system.
[0033]
Further, since the waveform estimation unit 20 accumulates a series of synthesized data until the end of the speech section and cuts out the recognition processing data after the end of the section, the waveform estimation unit 20 It is possible to extract recognition processing data having a predetermined number of bits that retains peak information included in the waveform, and the recognition rate can be increased.
[0034]
In the voice input device 100 of the present embodiment described above, the recognition processing data generation unit 28 in the waveform estimation unit 20 is the period synthesis data until the voice segment end determination unit 24 determines the end timing of the voice section. Since the recognition processing data is output after the accumulation period ends, a delay time corresponding to the accumulation period occurs. In order to shorten the delay time, for example, a short divided period may be set, and the recognition processing data may be output from the recognition processing data generation unit 28 for each divided period. However, if recognition processing data having a larger value than the recognition processing data output corresponding to the previous divided period is output, the recognition processing data in the subsequent divided period is cut out in the combined data. The position is changed, and the previous recognition processing data becomes invalid. At this time, it is necessary to notify the subsequent speech recognition apparatus 200 to interrupt the speech recognition processing and output again the recognition processing data whose cut-out position has been changed from the beginning. The subsequent speech recognition apparatus 200 performs speech recognition processing using the series of recognition processing data output again in this manner.
[0035]
In addition, when 16-bit recognition processing data is generated using the voice input device 100 according to the present embodiment described above, a recognition process in which voice data is not synthesized for an input voice signal having a low signal level. For the input voice signal having a large signal level, the voice data is synthesized to generate the recognition processing data. When the distortion (error) included in the recognition processing data increases due to the synthesis processing, it is desirable to change the recognition method in the speech recognition apparatus 200 according to the presence or absence of the synthesis processing.
[0036]
FIG. 3 is a diagram illustrating a modification of the voice input device and the voice recognition device. The voice input device 100A shown in FIG. 3 is different in that a level meter 30 is added to the voice input device 100 shown in FIG. The level meter 30 detects the signal level of the audio signal output from the amplifier 12. When this signal level exceeds a predetermined value, voice data synthesis processing is performed exceeding the input allowable voltage range of the analog-to-digital converter 14, so that the waveform is monitored by monitoring the detection output of the level meter 30. It is possible to determine whether or not the recognition processing data output from the estimation unit 20 is generated by the synthesis process.
[0037]
The speech recognition apparatus 200A shown in FIG. 3 includes a recognition processing unit 210, a normal acoustic dictionary 212, a distortion learning acoustic dictionary 214, and a switching unit 216. The normal acoustic dictionary 212 stores collation waveform data for recognizing the content of recognition processing data that has not been subjected to synthesis processing. The distortion learning acoustic dictionary 214 stores collation waveform data for recognizing the content of recognition processing data that has undergone synthesis processing. The switching unit 216 connects the normal acoustic dictionary 212 to the recognition processing unit 210 when the level value of the audio signal detected by the level meter 30 does not exceed the predetermined value, and when the level value exceeds the predetermined value, the distortion learning sound The dictionary 214 is connected to the recognition processing unit 210. The recognition processing unit 210 uses the normal acoustic dictionary 212 or the distortion learning acoustic dictionary 214 connected to the 16-bit recognition processing data output from the waveform estimation unit 20 in the speech input device 100A. Voice recognition processing is executed using waveform data.
[0038]
In this way, even if the same 16-bit recognition processing data is used, it is possible to further improve the recognition rate by switching the dictionary to be used depending on whether it is obtained by the synthesis processing.
[Second Embodiment]
FIG. 4 is a diagram illustrating a configuration of the voice input device according to the second embodiment. 4 includes a microphone 10, an amplifier 12, analog-to-digital converters (A / D) 14, 18, an attenuator 16, a waveform estimation unit 20B, and level meters 32, 34. It consists of This voice input device 100B is different from the voice input device 100 shown in FIG. 1 in that level meters 32 and 34 are added and the waveform estimation unit 20 is changed to a waveform estimation unit 20B.
[0039]
One level meter 32 detects the signal level of the audio signal input to one analog-digital converter 14. The other level meter 34 detects the signal level of the audio signal input to the other analog-digital converter 18. These level meters 32 and 34 correspond to level detecting means.
[0040]
The waveform estimation unit 20B synthesizes the two intermediate data output from the analog-digital converters 14 and 18 based on the level ratio between the two audio signals detected by the level meters 32 and 34.
In the first embodiment described above, the gain of the attenuator 16 is (1/2).¹⁵However, it is difficult to realize this gain accurately because there are actually variations in manufacturing of the elements constituting the attenuator 16. The waveform estimation unit 20B according to the present embodiment determines a bit position when combining the two intermediate data based on the level ratio of the two audio signals detected by the level meters 32 and 34. For example, the design gain of the attenuator 16 is (1/2)¹⁵If the level ratio (attenuation ratio) of the audio signal that matches this design value is ascertained from the detection results of the level meters 32 and 34 when it is set to double, it is output from the analog-digital converters 14 and 18. The two intermediate data are shifted by 15 bits to generate composite data. Further, from the detection results of the level meters 32 and 34, the actual gain of the attenuator 16 is (1/2).¹⁴When it is confirmed that the data is doubled, the two intermediate data output from the analog-digital converters 14 and 18 are shifted by 14 bits to generate composite data.
[0041]
As described above, in the voice input device 100B according to the present embodiment, it is possible to generate recognition processing data with less distortion by synthesizing voice data in consideration of manufacturing variations and the like. It becomes possible to increase the recognition rate of the speech recognition processing in the speech recognition apparatus 200.
[0042]
In addition, this invention is not limited to the said embodiment, A various deformation | transformation implementation is possible within the range of the summary of this invention. In the above-described embodiment, the voice input devices 100, 100A, and 100B are configured using two analog-digital converters. However, the voice input device is configured using three or more analog-digital converters. Also good.
[0043]
FIG. 5 is a diagram showing a configuration of a voice input device using three or more analog-digital converters. A voice input device 100C shown in FIG. 5 includes a microphone 10, an amplifier 12, an analog-digital converter 14, a plurality of attenuators 16, a plurality of analog-digital converters 18, and a waveform estimation unit 20C. For example, if the total number of analog-digital converters 14 and 18 is n and the bit length of data bits excluding the sign bits of the intermediate data output from each is m, the waveform estimation unit 20C has n × m bits. Composite data is generated by adding one sign bit to data bits, and recognition processing data having a predetermined number of bits is generated from the composite data. As described above, by increasing the number of analog-digital converters, it is possible to further widen the dynamic range when converting the input voice signal into the data for recognition processing. In addition, since each analog-digital converter can use an inexpensive one with a small number of bits, the cost of the entire apparatus can be reduced.
[0044]
In each of the embodiments described above, the number of bits of the intermediate data converted by each analog-digital converter is all the same, but analog-digital converters having different numbers of bits may be used in combination.
In each of the above-described embodiments, recognition processing data having a predetermined number of bits is extracted from the synthesized data having a large number of bits generated by the waveform estimation unit 20 or the like, but this synthesized data is extracted by the speech recognition apparatuses 200 and 200A. Can be processed as they are, the synthesized data itself may be output as recognition processing data. Even in such a case, it is possible to use an analog-digital converter having a small number of bits, and it is possible to reduce the cost while ensuring a wide dynamic range for the input audio signal.
[0045]
In each of the above-described embodiments, a plurality of audio signals having different amplitudes are generated using the attenuator 16. However, a plurality of audio signals having different amplitudes using an amplifier or a combination of the amplifier and the attenuator 16 are used. An audio signal may be generated.
[0046]
【The invention's effect】
As described above, according to the present invention, an encoding process in which a waveform is not partially lost even for an audio signal having a wide dynamic range in which the number of quantization bits of one analog-digital conversion unit is insufficient. In addition to ensuring a wide dynamic range, it is possible to increase the recognition rate of the speech recognition process by generating data including the entire speech waveform and inputting it to the speech recognition apparatus.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a configuration of a voice input device according to a first embodiment.
FIG. 2 is a diagram illustrating a detailed configuration of a waveform estimation unit.
FIG. 3 is a diagram showing a modification of the voice input device and the voice recognition device.
FIG. 4 is a diagram illustrating a configuration of a voice input device according to a second embodiment.
FIG. 5 is a diagram showing a configuration of a voice input device using three or more analog-digital converters.
[Explanation of symbols]
10 Microphone
12 Amplifier
14, 18 Analog-digital converter (A / D)
16 Attenuator
20, 20B, 20C Waveform estimation unit
22 Double precision data generator
24 Voice segment end determination unit
26 Valid data position monitoring unit
28 Recognition processing data generator
100, 100A, 100B, 100C Voice input device
200, 200A voice recognition device
210 Recognition processing unit
212 Normal acoustic dictionary
214 Distortion Learning Acoustic Dictionary
216 switching part

Claims (8)

A voice input device that is provided in a front stage of a voice recognition device and generates voice data corresponding to an input voice signal,
Signal generating means for generating a plurality of audio signals having different amplitudes by attenuating or amplifying the input audio signal with a predetermined gain ;
A plurality of analog-digital conversion means for converting each of the plurality of audio signals generated by the signal generation means into digital data;
Data combining means for combining the plurality of digital data output from the plurality of analog-digital conversion means;
Level detecting means for detecting the level of each of the plurality of audio signals generated by the signal generating means;
And the data synthesizing unit is configured to detect the bit positions of the plurality of digital data output from the plurality of analog-digital conversion units, and the levels of the plurality of audio signals detected by the level detection unit. A voice input device characterized by being synthesized by shifting by the number of bits corresponding to the ratio of. In claim 1,
A voice input device further comprising a microphone for collecting voice and outputting the input voice signal. In claim 1 or 2,
The number of the analog-digital conversion means is 2, and a stereo analog-digital converter is used. In any one of Claims 1-3,
The data synthesizing unit accumulates the digital data input in a predetermined period, and cuts out data of a predetermined number of bits including an audio signal having the largest amplitude in the period from the accumulated digital data. An audio input device that performs an output process. In claim 4,
The voice input device, wherein the predetermined period is a voice input section until the input voice signal is interrupted. In claim 4,
The predetermined period is a divided period shorter than the audio input period until the input audio signal is interrupted, and the amplitude of the audio signal corresponding to the subsequent divided period is the amplitude of the audio signal corresponding to the previous divided period If it is larger than the predetermined divided number of bits, the cut-out position of the data of a predetermined number of bits to be cut out is changed to a position corresponding to this large amplitude, and the process of cutting out and outputting from each accumulated digital data is performed from the previous divided period. A voice input device characterized by repetition. A speech recognition processing system comprising: the speech input device according to any one of claims 4 to 6 ; and a speech recognition device that performs speech recognition processing on data output from the speech input device,
For the digital data input during the predetermined period, the position for cutting out the data of the predetermined number of bits is fixed,
In the speech recognition apparatus, the data output from the data synthesizing means is generated using the digital data output from one analog-digital converting means, and the data is output from a plurality of analog-digital converting means. A plurality of acoustic dictionaries used for voice recognition processing depending on whether the digital data is generated using the digital data. In claim 7,
The plurality of acoustic dictionaries include a distortion learning acoustic dictionary that takes into account distortion that occurs when the digital data output from one analog-to-digital converter is combined, and a normal acoustic dictionary that does not consider this distortion. A speech recognition processing system characterized by comprising:

Images