Frequently Asked Questions List

Generally, when speaking politely and clearly, like a call center operator, the recognition rate is over 90-95% on average. On the other hand, when speaking casually, such as when speaking to a customer or on an internal line, the average rate is around 80-90%.

It is calculated by comparing what was actually said with the text of the speech recognition results.
If every single character matches, the recognition rate will be 100%. If characters are inserted, deleted, or replaced, the recognition rate will decrease.
For example, if "Osewa ni namasu" is recognized as "Osewa ni shite natte," then for the eight correct characters, one character is replaced by "su" to "shi" and one character is inserted, resulting in a recognition rate of (8-1-1) ÷ 8 = 75%.

Accurate speech recognition is difficult when the voice is quiet and unclear, difficult for the human ear to hear, when "arigatou gozashita" is abbreviated as "azashita," or when the voice contains strong noise.

Yes, it is possible. Technical terms and proper nouns can be recognized by voice recognition by "word registration." You can easily register words yourself.

Differences in intonation can be recognized without any problems. However, for words and phrases specific to dialects, such as saying "thank you" instead of "thank you," it is necessary to register the words and train the speech recognition engine.

Supported languages ​​will be announced separately. For details,Contact.

The following methods can be expected to improve recognition rates.

■ Word registration
Accuracy can be improved by registering technical terms and company terminology. Customers can easily register words themselves.

■ Individual tuning of the voice recognition engine
Our expert staff can improve the recognition rate by having the speech recognition engine learn the content of conversations at your contact center.

■ Operational improvements
You can improve the recognition rate by paying attention to the following:
Speak clearly and articulately
Set the recording volume appropriately
Maintain an appropriate distance between the headset and your mouth

Of the "acoustic model," "dictionary," and "language model," the "dictionary" can be registered by the user as a word registration function. By registering the spelling and reading (hiragana) of unrecognized words, speech recognition becomes possible.

Yes, sentiment analysis is implemented as standard.

Yes, we can do a proof of concept (POC) with a small number of seats. We can also provide you with pre-recorded voice data to test voice recognition. For details,Contact.

First, we will listen to your telephone environment and business operations, and then propose the optimal system configuration.
After that, the system will be installed and operation instructions will be provided before it goes into operation.
It usually takes a few weeks to a few months for the system to start up and running.

Yes, we offer a monthly fee service. For details,Contact.

Yes, it is possible. We will make the best proposal based on your budget and scale. For details,Contact.

DedicatedContactfrom the inquiry form.

A summary is automatically generated after the call ends. In addition, the call summary function allows administrators to summarize the contents of the call with the operator at any time.

Depending on your security requirements, you can choose between a local LLM that can be used in a local environment or a cloud-based generation AI. Local LLM does not require sending call data to an external cloud, and can be processed safely within the company's network. This reduces the risk of data leaks, even for highly confidential customer information and call content.

Yes, you can use cloud LLM or local LLM depending on the purpose, or you can combine the two for hybrid operation.

Yes, you can use existing recording data and a POC environment to preview the effectiveness of generative AI summarization and information extraction.Contact.

AmiVoice Communication Suite uses a speech recognition engine specialized for contact center conversations. Utilizing the extensive learning data accumulated through implementation in over 600 companies, it can convert call content into text with high accuracy.
It also supports individual tuning to suit your business operations and call characteristics, further improving speech recognition accuracy. It also supports the use of generative AI to summarize call content, generate Q&A, and automatically extract customer information. Depending on your security requirements, you can choose between local LLM, which can be used in a local environment, or the latest cloud LLM, and even combine the two for hybrid operation.

Yes, it is possible. If you can extract the audio file in the standard .wav format, it can be recognized.
However, recognition rate may decrease in cases of heavy compression or mono recording.

Voice recognition requires voice recording. We will propose the optimal recording method according to your telephone environment.

Yes, it is possible to integrate with systems via API. We have a variety of APIs available, allowing for integration for a wide range of purposes. Information such as call date and time, speaker information, and voice recognition results can be extracted by linking with external systems.

Yes, we can propose a system configuration that matches your thin client or telephone configuration. For details,Contact.

It supports both cloud and on-premise systems. You can also choose between real-time processing and batch processing. It can be flexibly constructed to suit your needs and environment. For details,Contact.

Yes, real-time and batch processing can coexist.

Yes, the recognition rate varies depending on the telephone environment and recording method.
We will propose the optimal configuration for your environment, so please start byAsk for.

Yes, we provide it as a cloud service. We can also build it into the cloud environment that you have contracted.