AmiVoice can be used for various purposes, fields, and environments.
It is compatible with many platforms including mobile phones and PCs.
It is compatible with both server-type and stand-alone type recognition methods.
Acoustic models include telephone speech, speech inside a car, and speech in noisy environments such as inside a factory, while language models include medical document creation, call center call records, and daily report creation, making it suitable for many industries and usage scenarios.

It can recognize anyone's voice. The recognition rate varies depending on the clarity of your pronunciation, speaking style, and the volume of your voice.

It accommodates differences in intonation and speed of speech. However, dialects that are not included in the dictionary will need to be newly registered.

AmiVoice outputs recognition results by looking at the preceding and following words, so if you break the utterance in the middle of a sentence, it may cause misrecognition. Try to speak in one sentence units as much as possible.

Use the tips below.

• Speak clearly and with a clear voice
• Speak in one sentence or punctuation mark
• Speak in multiple phrases

* When multiple phrases are entered, Japanese context analysis works more effectively.
*All sounds input from the microphone will be analyzed as Japanese sentences.
*Repeating something midway through or using words like "um" or "hmm" are considered words in context.

The most probable word sequence calculated using the acoustic model and linguistic model is used as the recognition result, and the recognition result, which includes both kanji and kana, is obtained directly from the speech. In speech recognition, hiragana is not recognized and then converted to kanji.

Recognition of individual hiragana characters is not in practical use because there are many instances of misrecognition.

Conversation is an engine that excels at natural speech between people, such as in face-to-face meetings, conferences, and web conferences.
Voice input is an engine that is strong at voice input to PCs and smartphones, such as voice operations and composing emails.

The general-purpose engine is a speech recognition engine that has been trained with a wide range of commonly used words and can be used for a variety of purposes.
Engines for each industry (domain-specific engines) such as medical, pharmaceutical, insurance, and finance are engines that are trained to specialize in the words unique to each industry, and are trained mainly on specialized terms and phrases within that industry. They can recognize industry-specific terms with high accuracy that cannot be recognized by general-purpose engines.

Recognition is processed in real time, and the recognition results are returned as soon as the speech is finished.
*Delays may occur due to network influences and congestion.
With synchronous and asynchronous HTTP speech recognition APIs, conversion to text will be completed within 1x the duration of the audio sent, at the latest.
For example, if you have an hour of audio, the text can be converted within an hour.
* In the case of the asynchronous HTTP speech recognition API, the speech recognition server is started separately after receiving a request, and it takes about one minute for the recognition process to start.

At AmiVoice API, our conversation engine learns new words every day.
Other engines are updated once a month.
*The engine does not learn for each user.

There are various possible causes for this, including environmental noise, the speaker's pronunciation, microphone performance, recording volume, and the engine not learning the words spoken.
If you feel the accuracy is poor, please check the recorded audio.
If the sound quality is good but the app is still misrecognizing your voice, try registering words.

Audio compressed with a lossy compression codec will be recognized less accurately than uncompressed audio.
The degree of degradation in recognition rate depends on the type of codec and compression rate.

AmiVoice API provides engines for Japanese, English, Chinese, and Korean, and a Thai engine is also available through AmiVoice API Private.
If you wish,CLICK HEREfrom the inquiry form.

It supports Mandarin Chinese and outputs recognition results in Simplified Chinese.

The differences and features are as follows:
　
1. Synchronous HTTP Speech Recognition API
It is suitable for uploading short audio files. The maximum audio data size that can be sent at one time is 16MB.
　
2. Asynchronous HTTP speech recognition API
Suitable for uploading long audio files. The maximum amount of audio data that can be sent at one time is 2.14GB. Even long audio files can be quickly converted to text. Speaker diarization function is available.
　
3. WebSocket Speech Recognition API
You can convert text into text in real time.
　
For more information,AmiVoice API Manual "Overview".

The AmiVoice API includes the written pronunciation, spoken pronunciation, speech start time, speech end time, and confidence.
For more information, please see the following page.
-Synchronous HTTP speech recognition API response
-For asynchronous HTTP speech recognition API responses

With the AmiVoice API, when using the WebSocket speech recognition API, the maximum time a session can be maintained is 24 hours.
Regardless of whether audio is being transmitted or not, if the session maintenance time has elapsed, the server will disconnect the connection. In this case, please reconnect.
For other restrictions,Limitationsplease confirm.

The AmiVoice API has a maximum data size that can be uploaded at one time: 16MB for synchronous HTTP speech recognition API, 2.14GB for asynchronous HTTP speech recognition API, and there is no limit for WebSocket speech recognition API.
For more information,Limitations.

The AmiVoice API has restrictions on the data format that you can upload.
For more information,Audio formatplease confirm.
Also, we do not support video formats. You will need to extract the audio data from the video yourself before sending it.

Generally, the format name must be specified, but there are exceptions where it can be omitted.
For more information,Audio format.

The AmiVoice API cannot be used in environments without an Internet connection.
AmiVoice SDK can be used without an internet connection. For more information,SDK .

With the AmiVoice API, when sending audio, you need to specify which engine to send it to.
For example, if you are using the conversation_general engine, specify "-a-general", and if you are using the English_general engine, specify "-a-general-en" after "grammarFileNames=" in the d parameter.
For a list of available connection engine names, seeMyPageIt is displayed in.
For more information,Connection engine name (grammarFileNames).

Aim to register around 1000 words per profile.
For more information, seeWord registration.

No problem.
However, please note that if multiple people use the service with one user ID, we will not be able to tally up the usage time of each user individually.

The AmiVoice API does not place a limit on the number of simultaneous connections, so you can send multiple audio streams at the same time using the same AppKey.
*Even if multiple audio files are sent at the same time, they will be recognized in parallel. However, delays may occur depending on the server's congestion, so please let us know in advance if you plan to send a large amount of audio at the same time.

AmiVoice API requires user registration even for trial use. When applying for use online (user registration),Terms of Use and SLA The contract will be completed upon your consent.
Each engine comes with 60 minutes free per month, so you can test it for free within that range.
To apply, CLICK HERE

CLICK HEREplease contact us using the Inquiry Form.
After verifying your identity, we will process the reissue.

You cannot change your user ID.

The email you sent may have been marked as spam by your environment.
The email for user registration will be sent from 'acp-info@amivoice.com', so please check your spam settings.
If you haven't received it, pleaseCLICK HEREfrom the inquiry form.

Terms of Use and SLA Anyone who agrees to the above can use this service.

MyPagePlease complete the cancellation procedure from the "Cancellation Procedure" on the website.

AmiVoice API is a pay-as-you-go system and does not offer a flat-rate plan.
However, if you are using the asynchronous HTTP speech recognition API for a long period of time, we may be able to accommodate flat-rate pay-per-use charges.CLICK HEREPlease contact us for more information.

With AmiVoice API, charges are only incurred for the parts of the audio data where voice is detected, that is, the speech sections that are the subject of voice recognition.
There is no charge for sections where no human voice is detected, however background music, television audio, voices from adjacent seats, and some other noises may be detected as speech segments.

With the AmiVoice API, telephone hold music is generally not subject to charges.
* Sections where guidance messages, etc. are recognized as human voices will be subject to charges.

The usage fee for 100 hours is as follows:
　
・AmiVoice API log saving enabled
For general-purpose engines (WebSocket/synchronous HTTP): 99 yen x 100 hours = 9,900 yen (tax included)
For general-purpose engine (asynchronous HTTP): 79.2 yen x 100 hours = 7,920 yen (tax included)
For finance and insurance industry engines: 118.8 yen x 100 hours = 11,880 yen (tax included)
For medical industry engines: 148.5 yen x 100 hours = 14,850 yen (tax included)
　
・AmiVoice API log not saved
For general-purpose engines (WebSocket/synchronous HTTP): 148.5 yen x 100 hours = 14,850 yen (tax included)
For general-purpose engine (asynchronous HTTP): 99 yen x 100 hours = 9,900 yen (tax included)
For finance and insurance industry engines: 148.5 yen x 100 hours = 14,850 yen (tax included)
For medical industry engines: 222.75 yen x 100 hours = 22,275 yen (tax included)
　
However, since you are only charged for the spoken portion, the actual amount is expected to be lower than the amount stated above.

For details on the fees,CLICK HEREPlease check the price list.

With the AmiVoice API, the speech start time (starttime) and speech end time (endtime) can be obtained for each speech section from the returned recognition results, so it is possible to aggregate the results by adding up the differences.
Incidentally,MyPageYou can also check this month's usage.
*Updated every 6 hours (please note that the XNUMX-hour interval may change).

There is no minimum usage fee per request for the AmiVoice API.

With the AmiVoice API, if your "usage" for the current month is within the free time range for each engine, or if the usage fee for each engine is less than 10 yen, you will not be charged.
* "Usage" is the cumulative time of the "speech section" of the audio data sent to the server (rounded down to the nearest second).

Logs refer to voice data and recognition results.
"Log storage" is offered at a lower price than "no log storage" if you agree to the use of these logs for research, development, and quality improvement of our products and services.
If you select "No Logs," voice data and recognition results will not be saved as files on the server. All processing will be done in memory, and they will be deleted from memory once processing is complete.
For details on when logs are left, please see the Q: "Please tell me the cases in which voice data and recognition results are stored on the server side."

1. Asynchronous HTTP speech recognition API
From the time the job is accepted until the recognition process is completed, the voice file is temporarily stored on the storage. This voice file is automatically deleted immediately after the recognition process is completed. The recognition results are also stored on the storage for one week and can be retrieved by the user. Access to these voice recognition results is protected by the AppKey and session ID.
　
2. When using the WebSocket speech recognition API or synchronous HTTP speech recognition API with "log storage" enabled
The voice data and the recognition result string are temporarily stored on the server, and then moved to a safe and robust storage by late-night batch processing. Some of this voice data may be used as material for machine learning in our voice recognition engine. Data that is not used for learning will be automatically deleted after a certain period of time.
　
3. When using the WebSocket speech recognition API or synchronous HTTP speech recognition API with "no logs saved"
Neither the voice data nor the recognition result string is stored on the server side. Everything is processed in memory, and once the recognition result is returned, it is deleted from memory.

Yes. All services are run on domestic servers.

AmiVoice API guarantees a monthly service uptime of 99.95% or higher.
Monthly service uptime = (total monthly operating time - cumulative outage time) ÷ total monthly operating time x 100
For more information,SLA .

AmiVoice API will notify you as soon as possible during the reception hours listed below on the login screen of the AmiVoice API service or on the ACP site.
Reception hours: 9:30AM to 5:30PM
(Excluding Saturdays, Sundays, national holidays, national holidays, and the year-end and New Year holidays designated by the company)
For more information,SLA .

With the AmiVoice API, voice recognition processing will remain available even during maintenance.
If we need to suspend the service and carry out maintenance, we will notify you at least three days in advance on the AmiVoice API service login screen or the AmiVoice Cloud Platform website.
*This does not apply in the event of an emergency.
For more information,SLA .

We have obtained the Privacy Mark.

There are three types of recognition: on-server recognition, on-device recognition, and hybrid recognition, which is used depending on the usage situation. When performing on-device recognition, recognition can be performed even if not connected to a network. When performing on-device recognition, the number of vocabulary words registered in the dictionary and language model depends on the device specifications.

Domain-specific engineIs available.

Yes, it is possible. For details and prices,CLICK HERE.

We can provide support in Japanese, English, Chinese, Korean, and Thai (as of March 2022).
The range of support for each language is different, soCLICK HERE.
*Packaged products are only available in Japanese, and multilingual support is available on an individual basis.

As for the price,CLICK HERE.

For examples,CLICK HERE.

to do so.

Yes, that is possible. If it is within the scope of the application you are developing, it can be used on multiple devices.

Once you have completed development, if you want to use the voice recognition app for your own company or sell it, you will need a "commercial license." The cost isCLICK HERE.

Not available.
This applies to desktop apps for Windows and native apps for Android and iOS.
*Please use AmiVoice API for voice recognition in web browsers.

AmiVoice SDK consists of libraries, documentation, and sample source code.

You can download it from the dedicated website.

We provide the AmiVoice API within your network.
Companies can build and operate their own voice recognition system suited to their environment, purpose, and service while complying with the security standards of their own company. This allows them to safely input voice data even in financial institutions, where information security is important, or for highly confidential corporate information that cannot be stored outside the company.

Rule grammarYou can use the grammar file stored on the server to perform recognition.

Domain-specific engineIs available.

Yes, it is possible. For details and prices,CLICK HERE.

As for the price,CLICK HERE.

We develop and build voice recognition IVR using Amazon Connect.

In addition to domain-specific speech recognition engines, you can customize recognition phrases to suit your IVR flow.

It does not support anything other than Amazon Connect.
If you want to develop a voice recognition IVR outside of Amazon Connect, you will need to develop it separately using the AmiVoice SDK.
CLICK HERE.

The audio files and voice recognition result files will be stored in the user's AWS S3.

For details on how to apply,CLICK HEREPlease contact us via the form below. After we have a consultation, you can start using the service after filling out the application form and completing the setup.

For information on fees and payment methods,CLICK HERE.

You will need to prepare an Amazon Connect environment on your side, as well as AWS Lambda, Kinesis Video Streams, Kinesis Data Streams, and Dynamo DB for voice recognition integration.

We provide documentation and sample programs. There is no significant difference in the skills required compared to standard development that does not use voice recognition.

We provide development support via email inquiries.

You can freely choose the AmiVoice speech recognition engine that is specialized for the content of call centers' calls.
* There are six selectable engines: general-purpose, finance, securities, communications, pharmaceuticals, and manufacturing.

You can register dictionary words by registering the reading and spelling of the word you want to register. You can also use a word list customized for voice recognition IVR.

You will be billed according to your monthly usage. As of April 2022, it is not possible to check your usage history each time.

This is a voice recognition server system that supports MRCP (Media Resource Control Protocol). It is possible to develop and build a voice recognition IVR that supports MRCP.

Both MRCP version 1 and MRCP version 2 are supported.
If your IVR supports the above, you can use the AmiVoice speech recognition engine by accessing the system from the IVR.

It does not correspond.

For details on how to apply,CLICK HERE.

For prices,CLICK HERE.

You will need to set up the entire AmiVoice MRCP Server system in a server environment that can connect to your MRCP system.

Development documentation is provided.
Scratch development is required on the MRCP IVR side.

We provide development support via email inquiries.

Various voice recognition engines are available.

You can register words from the system.
It is also possible to use rule grammars that are customized with original recognition phrases specific to IVR.

No