From September 9th, the standalone transcription system "AmiVoice MinutesWriter" has been updated to version 7. The main version updates are as follows.
【new function】
1. AmiVoice Recorder
■1-1. Revamped project startup method
Changes have been made so that project startup files can be managed all at once.
We have seen many cases where multiple startup project files (files with extensions avnrs/avsrs) are prepared, making the desktop cluttered. In order to make it easier to understand the relationship between project files and the voice recognition file "VText" and increase convenience, we have changed the system to centrally manage project files in the "Project List" within the system.
You can create a new project, rename the project, duplicate it, or delete it using the buttons on the right. You can also import previous project files using the "Import project" button. (Existing project files can be used even after this feature is released.)
If "Share settings information with all accounts" is set (default), the project list will be shared with the account that logs into the Recorder computer.
If you want to manage project registration for each login account, remove the check.
("Share settings information with all accounts" is set in "Operation Settings" in the "Settings" menu.)
■1-2. Added the ability to export VText files as one file without dividing them (feature available only in 1-bit version)
The option ``Automatically export a VText file after voice recognition is completed'' has been added to ``Recording data creation options''. If you select this, the VText file will not be split. However, it may take some time to finish recording, so please do not close the application until the message "Closing recorded data" disappears.
2. Common to AmiVoice Recorder/AmiVoice Recognizer
■2-1. Compatible with Bi-LSTM acoustic model
Now compatible with speech recognition engines that implement bidirectional deep learning technology "Bi-LSTM (Bidirectional Long Short-Term Memory)". This improves the error improvement rate and improves the recognition rate. We will gradually release engines that adopt Bi-LSTM.
■2-2. Enhancement of user word registration function
Improved user word management so that it can be registered in two types: "common word group" and "individual word group".
User words used for speech recognition
① No word registration
②Common word group
③Common word group + individual word group
You can choose from 3 patterns to use.
Each can be exported and imported to CSV files, making maintenance easy.
[Common word group registration screen]
[Individual word group registration screen]
Recorder Select and use registered words to use when recording.
3. AmiVoice Rewriter
■3-1. Addition of speaker labeling function (function when editing VText file)
When adding a speaker's name in Rewriter, we have added a function that "presumes the same speaker and automatically inputs the speaker's name".
This function uses the "features of the speaker's voice" in the selected line as training data, and automatically enters the speaker's name in the speaker name field after the selected line.
Accuracy is higher for longer utterances. Please put a ✅ in the longer part of the utterance.
However, there may be cases where the desired accuracy may not be achieved due to poor recording conditions, excessive speech fogging, etc. note that.
*Does not function when ControlManager is connected.
The speaker name is entered when it is determined that the speaker has changed in the line below the selected line.
■3-2. Addition of segment interlocking function during continuous playback
When checking the audio using the continuous playback function, I had to select the segment when I came to the segment I wanted to modify.
This time, we have added a function to automatically track segments, so when you reach the area to be corrected, please pause ("Ctrl" + "G") and edit the recognition results.
This function can be changed in ``Operation Settings'', ``Others'', ``Segment movement linked to cursor position''. If you want to use it with the same behavior as before, please change the setting to "No".
■3-3. Function added to “Export edited content”
Previously, when outputting multiple VText files to a single WORD file, it was necessary to select "Export Edited Contents" in the workspace view, and it was difficult to understand how to "Export Edited Contents". .
This time, we have revised this interface and added an area on the right side where you can select the VText to output when you click the "Export Edited Contents" button at the top of the screen with multiple VTexts open. The date and time are sorted in ascending order, so basically, just put a ✅ in the "Export all files into one file" box and start exporting.
If you want to change the print order, use the arrow keys on the right side of the screen to move up and down, and then output.
[Modification]
4. Common to AmiVoice Recorder/AmiVoice Recognizer
■4-1. Change the voice recognition setting to select between "accuracy priority" and "speed priority"
With the Bi-LSTM speech recognition engine, you can choose between "accuracy priority" and "speed priority". "Precision Priority" is the recommended setting that provides the best balance between processing time and recognition accuracy. If you want to speed up the processing time, please use "Speed Priority".
Also, since the interfaces for Recorder and Recognizer were different, we unified them into the same interface.
■4-2. “Utterance detection settings” has been abolished.
The ``segmenter'' that detects utterances has been designed to use the latest utterance detection module. The settings screen has been abolished as it is no longer necessary.
5. Recommended environment
■5-1. About Windows7 Pro
Windows 7 Pro, which is no longer supported by Microsoft, has been excluded from the list of confirmed operating systems. Since many companies and local governments have already stopped using Windows 7, we have excluded Windows 7, which is no longer supported, from the operation confirmation test.
■5-2. About recommended OS
Since memory usage tends to increase due to annual improvements, the recommended OS is the 64-bit version of Windows OS. As deep learning-based speech recognition engines have evolved, the number of vocabulary words in dictionaries has gradually increased, and memory consumption has also increased year by year.
Based on the prediction that memory usage will not decrease in the future, we will recommend a 64-bit version of Windows OS for new installations in the future.
6. Other
■6-1. Update HASP driver to SentinelRuntime Ver8.11
Compatible with Windows 10 Pro Ver2004.
Due to an issue with Windows 10's "Windows 10 May 2020 Update" (version 2004), which is being provided by Microsoft, it has been reported that the USB key used to start the editing software "AmiVoice Rewriter" cannot be used. Masu.
Before updating to Windows 10 Pro Ver. 2004, please update with HASP driver Ver. 6.71 stored in the application CD, version 8.11.
