Thorough verification of the accuracy of the AI voice recorder "Automemo" of the source nextenext that can automatically transcribe characters

■ Serialization / Junya Ishino's Gachi Review

　A new IoT device has appeared from the source nexust, which made a more compact translator "Pocketalk" than a smartphone.That is the AI voice recorder "Automemo".At first glance, it is a slightly large voice recorder, but the feature of this product is automatic transcript of characters in cooperation with the cloud.

　Automemo automatically raises audio files on the cloud when connecting to Wi-Fi and takes a text.It will be easier to search for the required audio files in the search later, and to regenerate only places you want to hear.The audio file is automatically uploaded when connected to Wi-Fi, and simply referring to it from the application saves the trouble of transferring files to PCs, unlike conventional voice recorders.

　Because it uses a subscription -type business model, there is no need to worry about the rest of the capacity.The price is 980 yen per month for "Premier Plan", which can be used for recording data until 30 hours a month.If you pay an additional 980 yen, you can add time to make text in 10 hours.If it is free, the texture function can be used for 1 hour a month.So what are the actual usability and the accuracy of writing characters?Verified the ability with the actual machine.

AUTOMEMO of the source nexust that was called AI voice recorder

Super simple operability excluding even displays

　Automemo looks very simple than other voice recorders.There are no generally installed displays, and the buttons are simple with two in front.Since there is no playback function, there is no volume adjustment button.The terminals on the bottom are not for earphones, but to connect the microphone.Because it is a device that is divisible to the last, the design reflects that.

The appearance is very simple, only two buttons on the front

There is a hole like an earphone jack on the bottom, which is for microphone

Equipped with a power key on the side

　The operation is so easy.Once you connect to Wi-Fi with the initial setting, just turn on the power and press the button to start recording.To stop recording, just press the same button again.During the recording, the LED lights up and it is designed to understand it.As mentioned above, it is difficult to understand which state is a bit difficult to understand now because there is no display, but it seems to be a trade -off because the buttons will increase and become complicated.

　The initial setting requires a smartphone, but there is no difficult operation.When Automemo is connected to a charging cable and the smartphone app is activated, Automemo is automatically detected.In this state, you can select the Wi-Fi SSID you want to connect and enter the key.When used, there is no need to be connected to Wi-Fi at all times, and after recording, when you enter the Wi-Fi area, the audio data is automatically uploaded.After uploading, after the texture is over, it will be notified to the set e -mail address and smartphone.I recorded a press conference for about an hour, but the text was completed without waiting so much.

Connect from a smartphone app when charging to make the initial settings

　The slightly small button under the recording button is for bookmarking.If you push it small at the time of recording, it will be a line break there, and it will be easier to understand the separation of the remarks later.Conversely, without a bookmark, all the texts will continue, making it difficult to decipher later.It's a bit of a hassle, but to make effective use of the texture function, it's better to use the bookmark function properly.

The accuracy of texture varies greatly depending on the environment, is it not good at colloquial?

　So what is the accuracy of the essential texture?The following screenshots are textbooks at the beginning of Rakuten's financial results.Because it was an online conference, Automemo was placed near the speaker.Regarding sound, it is better than a realistic press conference, and it is an environment that is "easy for AI".As you can see, it is relatively accurate, but there are some places where recognition is sweet.

I tried to convert the online conference.The accuracy is reasonable, but there are mistakes in some places

　More than that, I was worried that it was difficult to understand where the remarks were separated because the punctuation was not attached, and I was worried that the readability was reduced.Certainly, it is difficult for the speaker to make a reading point, but the phrase can be used automatically when a certain period is available.I felt that there was room for reconsideration regarding the readability of text -ized voice.

　Also, as in the text that reads the manuscript, I was worried that the speech was relatively accurate for the audio that was established as a sentence, but the accuracy would decrease when it became a spoken language.For example, even in the same conference, the accuracy is decreasing when you look at the texts of scenes that do not have a pre -writing answer, such as questions and answers.There are many places where the awareness was not well done, and there were many places where the middle was falling off, so it is difficult to see at a glance which part has been turned into text.

If you have a lot of ad -lib exchanges in colloquial, it often becomes a meaningless text.

　In fact, while reading out the manuscript that has become a sentence and putting on a bookmark in each paragraph, it is quite accurate.Looking at the results, it is strong in literary language, but the colloquial is not good.However, in this state, it is convenient to be able to search for the contents of the audio file with keywords.In the case of the author, if you search for keywords such as "5G" or "price plan" or in proper nouns in the company name, the target audio will be displayed in one shot, so open the contents when you want to re -listen later.No need to check.

While reading the manuscript and putting on the bookmark, the accuracy was relatively improved.

　It may be a disappointing result for the direction of the expectation that it will be a beautiful sentence, but it is said that Japanese is especially difficult to make text.The difference between the spoken language and the literary language is large, and there are many cases where the subject is omitted and conversations that ignore the grammar are large, so the hurdles are likely to be higher than the language of English.It is better to assume that you do not expect excessive expectations and make it easier to search for recorded data later.

Expectations for improving accuracy even for parts that are not enough in terms of functionality

　Automemo is simple and does not get lost in operation, but there is room for improvement in terms of usability.The corresponding frequency of Wi-Fi is one of them.Is it because of cost reduction?.It can only be used in the 4GHz band, but it is a pity that it is not compatible with the 5GHz band, which is harder to interfere at faster.In consideration of radio interference and communication speed, I basically use only 5GHz bands in the office.Therefore, 2.To connect Automemo in the 4GHz band to the Internet, it was necessary to use smartphone tethering.

The corresponding frequency of Wi-Fi is 2.Only 4GHz band.Not supported by mobile data communication

　欲を言えば、ソースネクストのPOCKETALKのように、モバイルデータ通信にも対応していてほしかった。テザリングで利用すれば、それに近いことはできる一方で、やはり機能をオンにするひと手間がかかる。モバイルデータ通信があれば、外出先で音声を録音して、その場ですぐにアップロード&テキスト化したデータを、自宅や事務所などのWi-Fiがある場所に戻る前に確認できる。既存のボイスレコーダーは、その場でサッと確認できたため、それに近いことをしようとすると、やはりモバイルデータ通信が必要だ。

　Also, I want you to access Automemo sites from your PC as well as apps.Checking the recorded sounds and text is rather when sitting in front of the desk and sitting down and writing manuscripts.It is convenient to be able to confirm while moving with the app, but it is a little troublesome to check your smartphone during the desk work.If possible, I want you to be able to refer to not only the app but also from the web.

Apps are required to check the recorded voice and text

　Automemo has the impression that there is still a rough place, but the concept of voice recorders that can make voice textbooks is revolutionary.It seems that the same thing can be done with a smartphone app, but the advantage of the division of devices is great.If it is a smartphone, it is difficult to perform other operations during recording and cannot respond to incoming calls.The corresponding frequency of Wi-Fi and the correspondence to mobile data communication depends on the hardware, so you can only expect the successor, but the accuracy of texture can be updated on the cloud side, so it is continuous.I want to look forward to evolution.

[Ishino's Judgment] Easy to hold ★★★★★ UI ★★★★ Connection performance ★★ Texture accuracy ★★★ Battery mochi ★★★★ * Scoring is judged with 5 points in each item

Coverage / sentence / Junya Ishino

After graduating from Keio University, joined Takarajimasha.After independence, he has been active in a wide range of media as a mobile journalist/writer.He has authored many books, including "Catayildren" (Softbank Shinsho) and "Easy to Understand in 1 hour" (Mainichi Shimbun).