CN105741834A - Voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation - Google Patents
Voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation Download PDFInfo
- Publication number
- CN105741834A CN105741834A CN201410753671.5A CN201410753671A CN105741834A CN 105741834 A CN105741834 A CN 105741834A CN 201410753671 A CN201410753671 A CN 201410753671A CN 105741834 A CN105741834 A CN 105741834A
- Authority
- CN
- China
- Prior art keywords
- recording
- frequency spectrum
- outfit
- sound pick
- voice recording
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The invention discloses a voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation. The method comprises: step a, the same kind of common voice recording equipment like a recording pen is selected and a same-kind voice recording equipment base is constructed, wherein the number of the voice recording equipment is n; step b, voice recording equipment claimed by a to-be-checked recorded voice, a voice recording environment, and a recording parameter are obtained and the voice recording equipment claimed by the to-be-checked recorded voice is added to the voice recording equipment base; step c, the voice recording equipment in the voice recording equipment base is used for carrying out experiment voice recording in a similar recording environment and on a condition of the recording parameter, wherein the voice recording time is larger than or equal to t; step d, frequency spectrum statistic feature calculation is carried out on the to-be-checked recorded voice and the experiment voice recorded by the step c; step e, classifier training for equipment classification in the voice recording equipment base is carried out experiment voice spectrum statistic features calculated by the step d by using a support vector machine method; step f, spectrum statistic features, calculated by the step d, of the to-be-checked recorded voice are classified by using the classifier calculated by the step e; and step g, statistic calculation is carried out on a classification result obtained by the step f and the voice recording equipment of the to-be-checked recorded voice is identified and verified according to a classification result criterion. According to the invention, the method has the following beneficial effects: a voice recording equipment identification problem is solved by combining the same-kind recording equipment base construction from the perspective of frequency spectrum statistic calculation of the voice recording signal in a similar environment; and an accurate and scientific judicial evidence taking and authentication application technical method can be provided.
Description
Technical field
The present invention relates to and calculate according to the frequency spectrum statistical nature of recording, and artificial intelligence machine learning method, rely on similar sound pick-up outfit storehouse reference, start with from the profound characteristic angle analysis of recorded audio signals, solve for the sound pick-up outfit identification problem judging this recording institute recording arrangement according to recording, to reach more accurate and more scientific judicial evidence collection and to identify application.
Background technology
Along with audiovisuals becomes China's judicial evidence form, recording material is as the important component part of audiovisuals, and the development of its verity judicial evidence collection and new Identification technology new method has important theoretical significance and actual application value.Sound pick-up outfit identification for recording is the important channel of checking recording verity and checks angle.Traditional confirm that the method that the sound pick-up outfit whether recording to be checked is claimed is recorded has in actual applications based on methods such as audition inspection and metadata analysises and be greatly limited, be in particular in that accuracy is not high, and cannot effectively check the situation distorting forgery.Judge that the sound pick-up outfit identification problem of this recording institute recording arrangement is badly in need of starting with from deeper level characteristic angle analysis according to recording.This patent is complied with under this situation demand exactly, frequency spectrum statistical nature calculating from recording, and combination supporting vector machine machine learning method, rely on similar recording environment and similar sound pick-up outfit storehouse, it is achieved the sound pick-up outfit identification of frequency spectrum statistical computation of recording based on similar environments.
Summary of the invention
For designing judicial evidence collection and the identification technology of more accurate and science, the present invention provides a kind of foundation similar environments recording frequency spectrum statistical nature to calculate the sound pick-up outfit recognition methods learning with support vector machine and classifying.
This invention address that the technological means that technical problem adopts is:
Based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation, wherein, comprising the steps: step a: select common similar sound pick-up outfit, such as recording pen etc., build similar sound pick-up outfit storehouse, sound pick-up outfit quantity is n；
Step b: understand recording arrangement, playback environ-ment and recording parameter that recording to be checked is claimed, the recording arrangement that recording to be checked is claimed adds in sound pick-up outfit storehouse；
Step c: using the sound pick-up outfit in sound pick-up outfit storehouse at similar recording environment and to record recording experiment recording under Parameter Conditions, long recording time is be more than or equal to t；
Step d: treat the experiment recording recorded in call the roll of the contestants in athletic events sound and step c and carry out frequency spectrum statistical nature calculating；
Step e: use support vector machine method to carry out the experiment recording frequency spectrum statistical nature calculated in step d for the classifier training of device class in sound pick-up outfit storehouse；
Step f: use the step e grader calculated that the recording frequency spectrum statistical nature to be checked calculated in step d is classified；
Step g: the classification results of step f is carried out statistical computation, according to classification results criterion identification and the recording arrangement verifying recording to be checked.
The sound pick-up outfit recognition methods of above-mentioned frequency spectrum statistical computation of recording based on similar environments, wherein, the frequency spectrum statistical nature computational methods in described step d are as follows:
(1) the background noise recording in recording is extracted according to fixed threshold method.
(2) background noise recording is carried out equal length zero overlapped partitioning.
(3) the background noise recording of every section of segmentation is carried out discrete Fourier transform calculating, and the Fourier Transform Coefficients calculating gained is normalized calculating, obtain the frequency spectrum statistical nature of this segmentation background noise recording.
(4) set that the frequency spectrum statistical nature recorded is is elementary cell with the frequency spectrum statistical nature of every section of background noise recording.
The sound pick-up outfit recognition methods of above-mentioned frequency spectrum statistical computation of recording based on similar environments, wherein, the classifier training method in described step e is as follows:
(1) frequency spectrum statistical nature and its classification logotype to the experiment recording in sound pick-up outfit storehouse carry out labelling；
(2) use support vector machine method that frequency spectrum statistical nature and its classification logotype of the experiment recording in sound pick-up outfit storehouse are carried out classifier training.
The sound pick-up outfit recognition methods of above-mentioned frequency spectrum statistical computation of recording based on similar environments, wherein, the recording frequency spectrum statistical nature sorting technique to be checked in described step f is as follows:
(1) use the grader in step e to treat each section of background noise recording frequency spectrum statistical nature called the roll of the contestants in athletic events in sound spectrum statistical nature to classify；
(2) classification results of all background noises recording frequency spectrum statistical nature constitutes recording frequency spectrum statistical nature classification results to be checked.
The sound pick-up outfit recognition methods of above-mentioned frequency spectrum statistical computation of recording based on similar environments, wherein, the classification results criterion in described step g is as follows:
(1) classification results in step f is carried out number statistics by class categories；
(2) percentage ratio that there is the class categories of maximum statistical number in all class categories numbers is calculated, if percentage ratio exceedes threshold value t, then it is assumed that recording to be checked is recorded by the sound pick-up outfit having represented by the class categories of maximum statistical number.
The invention has the beneficial effects as follows:
(1) present invention devises the sound pick-up outfit recognition methods based on similar environments recording and similar sound pick-up outfit storehouse reference.
(2) present invention devises and calculates the sound pick-up outfit recognition methods with support vector machine machine learning and classification according to recording frequency spectrum statistical nature.
(3) present invention devises the sound pick-up outfit recognition methods according to recording background noise fragmented spectrum tagsort result Statistical Decision Criterion to be checked.
Accompanying drawing explanation
Fig. 1 is the present invention flow chart based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation.
Detailed description of the invention
Below in conjunction with concrete Application Example, the invention will be further described, but not as limiting to the invention.This application embodiment provides the sound pick-up outfit recognition methods in judicial expertise application, and recording to be checked is wav form, and signal sampling rate is 32KHz, claims and is recorded in quiet office environment by PHILIPSSA5MXX04RFC recording pen, long recording time 30 minutes.This application embodiment comprises the steps: step a: select 20 recording pens common on the market, adds the sound pick-up outfit storehouse of PHILIPSSA5MXX04RFC recording pen composition totally 21 recording pens claimed, device name is as shown in table 1.
Table 1 sound pick-up outfit storehouse
Step b: using 21 recording pens in quietly office indoor recording experiment recording, duration is 1 hours.
Step c: calculating recording to be checked and the frequency spectrum statistical nature of experiment recording, the frequency spectrum statistical nature concrete grammar of recording is as follows:
(1) the background noise recording in recording is extracted according to fixed threshold 0.15.
(2) background noise recording being carried out equal length zero overlapped partitioning, every section of duration is 3 seconds.
(3) the background noise recording of every section of segmentation is carried out discrete Fourier transform calculating, window size is 0.1 second, Fourier's factor is sized to 2048, and 1024 Fourier Transform Coefficients calculating gained are normalized calculating, obtains the frequency spectrum statistical nature of this segmentation background noise recording.
(4) set that the frequency spectrum statistical nature recorded is is elementary cell with the frequency spectrum statistical nature of every section of background noise recording.
Step d: use the frequency spectrum statistical nature that the experiment that the sound pick-up outfit in 21 sound pick-up outfit storehouses is recorded by support vector machine method is recorded to carry out classifier training；
Step e: using the step d grader calculated to treat sound spectrum statistical nature of calling the roll of the contestants in athletic events and classify, the Characteristic Number participating in classification is 434；
Step f: classification results is carried out statistical computation, it is 428 that recording frequency spectrum statistical nature to be checked is categorized as the number of PHILIPSSA5MXX04RFC recording pen recording, accounting for always ratio 98.62%, more than the threshold value 80% set, result thinks that recording to be checked is recorded by PHILIPSSA5MXX04RFC recording pen.
The foregoing is only the present invention a Application Example, not thereby limit the claim of the present invention, so the equivalent structure change done by all utilizations description of the present invention and diagramatic content, be all contained in protection scope of the present invention.
Claims (5)
1. based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation, it is characterised in that comprise the steps:
Step a: select common similar sound pick-up outfit, such as recording pen etc., builds similar sound pick-up outfit storehouse, and sound pick-up outfit quantity is n；
Step b: understand recording arrangement, playback environ-ment and recording parameter that recording to be checked is claimed, the recording arrangement that recording to be checked is claimed adds in sound pick-up outfit storehouse；
Step c: using the sound pick-up outfit in sound pick-up outfit storehouse at similar recording environment and to record recording experiment recording under Parameter Conditions, long recording time is be more than or equal to t；
Step d: treat the experiment recording recorded in call the roll of the contestants in athletic events sound and step c and carry out frequency spectrum statistical nature calculating；
Step e: use support vector machine method to carry out the experiment recording frequency spectrum statistical nature calculated in step d for the classifier training of device class in sound pick-up outfit storehouse；
Step f: use the step e grader calculated that the recording frequency spectrum statistical nature to be checked calculated in step d is classified；
Step g: the classification results of step f is carried out statistical computation, according to classification results criterion identification and the recording arrangement verifying recording to be checked.
2. as claimed in claim 1 based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation, it is characterised in that the frequency spectrum statistical nature computational methods in described step d are as follows:
(1) the background noise recording in recording is extracted according to fixed threshold method；
(2) background noise recording is carried out equal length zero overlapped partitioning；
(3) the background noise recording of every section of segmentation is carried out discrete Fourier transform calculating, and the Fourier Transform Coefficients calculating gained is normalized calculating, obtain the frequency spectrum statistical nature of this segmentation background noise recording；
(4) set that the frequency spectrum statistical nature recorded is is elementary cell with the frequency spectrum statistical nature of every section of background noise recording.
3. as claimed in claim 1 based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation, it is characterised in that the classifier training method in described step e is as follows:
(1) frequency spectrum statistical nature and its classification logotype to the experiment recording in sound pick-up outfit storehouse carry out labelling；
(2) use support vector machine method that frequency spectrum statistical nature and its classification logotype of the experiment recording in sound pick-up outfit storehouse are carried out classifier training.
4. as claimed in claim 1 based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation, it is characterised in that the recording frequency spectrum statistical nature sorting technique to be checked in described step f is as follows:
(1) use the grader in step e to treat each section of background noise recording frequency spectrum statistical nature called the roll of the contestants in athletic events in sound spectrum statistical nature to classify；
(2) classification results of all background noises recording frequency spectrum statistical nature constitutes recording frequency spectrum statistical nature classification results to be checked.
5. as claimed in claim 1 based on the sound pick-up outfit recognition methods of similar environments recording frequency spectrum statistical computation, it is characterised in that the classification results criterion in described step g is as follows:
(1) classification results in step f is carried out number statistics by class categories；
(2) percentage ratio that there is the class categories of maximum statistical number in all class categories numbers is calculated, if percentage ratio exceedes threshold value t, then it is assumed that recording to be checked is recorded by the sound pick-up outfit having represented by the class categories of maximum statistical number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410753671.5A CN105741834A (en) | 2014-12-10 | 2014-12-10 | Voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410753671.5A CN105741834A (en) | 2014-12-10 | 2014-12-10 | Voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105741834A true CN105741834A (en) | 2016-07-06 |
Family
ID=56239994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410753671.5A Pending CN105741834A (en) | 2014-12-10 | 2014-12-10 | Voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105741834A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109935234A (en) * | 2019-02-22 | 2019-06-25 | 东莞理工学院 | A kind of method of pair of recording identification source device |
| CN110189767A (en) * | 2019-04-30 | 2019-08-30 | 上海大学 | A kind of recording mobile device detection method based on dual-channel audio |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102779281A (en) * | 2012-06-25 | 2012-11-14 | 同济大学 | Vehicle type identification method based on support vector machine and used for earth inductor |
-
2014
- 2014-12-10 CN CN201410753671.5A patent/CN105741834A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102779281A (en) * | 2012-06-25 | 2012-11-14 | 同济大学 | Vehicle type identification method based on support vector machine and used for earth inductor |
Non-Patent Citations (1)
| Title |
|---|
| 曾锦华等: "录音设备识别司法鉴定技术研究", 《中国司法鉴定》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109935234A (en) * | 2019-02-22 | 2019-06-25 | 东莞理工学院 | A kind of method of pair of recording identification source device |
| CN110189767A (en) * | 2019-04-30 | 2019-08-30 | 上海大学 | A kind of recording mobile device detection method based on dual-channel audio |
| CN110189767B (en) * | 2019-04-30 | 2022-05-03 | 上海大学 | Recording mobile equipment detection method based on dual-channel audio |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Nagrani et al. | Voxceleb: a large-scale speaker identification dataset | |
| Neal et al. | Time-frequency segmentation of bird song in noisy acoustic environments | |
| CN111279414B (en) | Segmentation-based feature extraction for sound scene classification | |
| JP2005530214A (en) | Mega speaker identification (ID) system and method corresponding to its purpose | |
| CN107507626B (en) | Mobile phone source identification method based on voice frequency spectrum fusion characteristics | |
| CN107609149B (en) | Video positioning method and device | |
| CN102394062A (en) | Method and system for automatically identifying voice recording equipment source | |
| Korshunov et al. | Cross-database evaluation of audio-based spoofing detection systems | |
| Liu et al. | Fast copy-move detection of digital audio | |
| US11653062B2 (en) | Methods and apparatus to determine audio source impact on an audience of media | |
| Cotton et al. | Soundtrack classification by transient events | |
| Gallardo-Antolín et al. | Histogram equalization-based features for speech, music, and song discrimination | |
| Ntalampiras et al. | Automatic recognition of urban soundscenes | |
| CN103594083A (en) | Technology of television program automatic identification through television accompanying sound | |
| US20090055336A1 (en) | System and method for classifying multimedia data | |
| CN105741834A (en) | Voice recording equipment identification method based on similar environment voice recording spectrum statistics calculation | |
| CN107274912B (en) | Method for identifying equipment source of mobile phone recording | |
| Lu et al. | Unsupervised speaker segmentation and tracking in real-time audio content analysis | |
| CN102789780B (en) | Method for identifying environment sound events based on time spectrum amplitude scaling vectors | |
| CN103440868B (en) | A kind of authentication method of electronics modified tone audio frequency | |
| Verma et al. | Cell-phone identification from recompressed audio recordings | |
| CN109935234B (en) | Method for identifying source equipment of sound recording | |
| Bestagini et al. | Feature-based classification for audio bootlegs detection | |
| Jiqing et al. | Sports audio classification based on MFCC and GMM | |
| Jørgensen et al. | Unsupervised speaker change detection for broadcast news segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 200063 No. 1347 Guangfu West Road, Shanghai, Putuo District Applicant after: ACADEMY OF FORENSIC SCIENCE Address before: 200063 No. 1347 Guangfu West Road, Shanghai, Putuo District Applicant before: INSTITUTE OF FORENSIC SCIENCE, MINISTRY OF JUSTICE PRC |
|
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160706 |