Batch Video and Audio Prohibited Words Detection

Batch-identify local audio and video content, automatically generate timeline subtitles, and scan them against a prohibited words list. When matches are found, a detailed report is output for easy review and correction.

Bug report

Loading tool, please wait...

If the online tool fails to load or run, try the desktop tool.https://tools.yikeaigc.com/

Tool Usage

Return to old version

A desktop version is available with more powerful, stable features. Open desktop version

Instructions

Software Usage Instructions

Select files to scan: Click the "Browse Files" or "Browse Folder" button to add the audio/video files or folders to be scanned. Multiple selection and drag-and-drop are supported. Separate multiple paths with English semicolons. Common formats such as MP4, AVI, MKV, MOV, FLV, MP3, WAV, FLAC, and M4A are supported.
Set traversal method:
- Traverse subdirectories: When checked, the audio/video files in the main folder and all its subdirectories will be processed recursively
- Keep original path structure: When checked, the output reports will follow the same directory structure as the source folder, and serial numbers will be automatically appended for duplicate names
Select prohibited words file: By default, the program reads "违禁词.txt" in the same directory as the program. Enter one keyword per line. UTF-8 and GBK encoding are supported, and lines starting with # are treated as comments. You can also click the "Browse" button to specify another text file.
Configure detection options:
- Report format: TXT or CSV output is available, recording the index number, start time, end time, subtitle content, and matched word
- Generate matched reports only: When checked, no report will be generated for files that do not contain prohibited words
- Keep intermediate subtitles: Also saves the recognized SRT subtitles for convenient manual review
- Maximum VAD segment length in milliseconds: Adjusts the splitting threshold for long sentences; it can be increased appropriately for long videos
- Device and threads: Choose CPU or CUDA GPU inference and set the number of parallel threads
Select save directory: Set the output location for detection reports and subtitle files. Drag-and-drop is supported.
Start detection: Click the "Start Detection" button to launch the batch task. The progress bar and log below will display the current file, recognition progress, and match status in real time. You can click the "Stop" button to interrupt processing.
View reports: After detection is complete, go to the save directory to view the TXT or CSV reports, locate where prohibited words appear on the timeline, and perform secondary editing or modifications.

FAQ

A: For video, formats such as MP4, AVI, MKV, MOV, FLV, WMV, WEBM, M4V, MPG, TS, 3GP, and RMVB are supported; for audio, formats such as MP3, WAV, FLAC, AAC, OGG, M4A, WMA, OPUS, AC3, and AMR are supported. The recognition engine is optimized for Mandarin Chinese content. Accuracy may decrease for dialects or materials with accents.

A: The prohibited words file is a plain TXT text file, with one keyword per line. Blank lines and comment lines starting with # are automatically ignored. It is recommended to save it in UTF-8 encoding, and GBK is also supported. You can group entries by topic and add comment lines to make it easier to maintain multiple lists such as industry-sensitive words, brand terms, and compliance terms.

A: Each matched record includes five fields: number, start time, end time, subtitle content, and matched prohibited word. TXT is suitable for reading manually and forwarding by email, while CSV can be opened in Excel or WPS for filtering and statistics, making it easier for teams to collaborate on corrections.

A: No. Speech recognition is completed using the local FunASR model, and prohibited word matching is also performed in local memory. The entire process does not connect to the internet or upload any audio, video, or subtitle content, making it suitable for processing materials involving privacy or trade secrets.

A: You can optimize it in the following ways: 1) If your device has an NVIDIA graphics card and the CUDA version of PyTorch is installed, switch the device to GPU; 2) appropriately increase the maximum VAD segment length in milliseconds to reduce splitting overhead; 3) check "Generate hit report only" to reduce IO; 4) add folders for processing in batches to avoid too many files in a single task.

A: You can use the hotword feature to improve recognition accuracy for proper nouns, personal names, and brand names. It is also recommended to include common homophones, phonetic variants, and abbreviations of keywords in the prohibited word list, and use the option to keep SRT subtitles for manual review to improve overall coverage.

A: It is suitable for compliance self-checks before publishing short videos on self-media platforms, review of e-commerce livestream clips, content review for training courses and audiobooks, sensitive word screening for ad materials, bulk correction of historical assets, and other scenarios, helping content teams identify risks in advance before uploading.

A: This tool focuses on detection and locating, and outputs an accurate timeline for each matched segment. After receiving the report, you can use video editing software or this site's video editing and voiceover replacement tools based on the timestamps for further processing, such as deleting, replacing, or muting the content.