Patent technology provider：Panasonic Co., Ltd.
「Free trial of the Voice Emotion Recognition」started in Nov. 2017 will be terminated on Aug. 21, 2018.
During the period, we thank you very much for your usages, feedback, and suggestions.
We are under development of the improved edition for your usage in this service. Please do not miss it.
"Human emotion is invisible" - [Emotion Recognition technology] makes it possible. This technology reads the human emotion through facial expression, text, conversations, and voice. In recent years together with the voice recognition technology, practical realization of the emotion recognition technology by voice is moving forward like installing the emotion recognition technology into robot.
Our emotion recognition technology determine the emotions by using voice / conversations.
The Voice Emotion Recognition technology is divided to [the language analysis type] and [the acoustic analysis type]. For [the language analysis type] which analyzes emotions through speech recognition of words had difficulty to determine emotions in cases of reading words expressing multiple emotions and containing multiple proper nouns, but our technology can read emotions without influences of those problems by using [the acoustic analysis type] based on the feature values of voice.
This enables to detect claims / complains which were difficult to determine by only [the language analysis type] and angry voices which were hard for speech recognition.
~Differences of language analysis and acoustic analysis~
The engine specification with energy saving and light operation can analyze conversation in almost real time. It can determine the emotions immediately (about 300ms) after voice input. (this is subject to change depending on the usage environment.)
For example, in call center when any claim (anger/stress) occurs which require the follow up by supervisor, this can correspond appropriately and smoothly without missing the timing.
~ Installation example 1. Call center (customer service) ~
Other than the real time determination of emotions, it is also possible to analyze voice and conversations after the batch processing of them. Installing to internal interviews as an example, prevention of power harassment by checking if the interview was done appropriately through the feedback in human resource, and plan the decline of cancellation rate, the improvement of customer satisfaction rate, by analyzing customer conversations and emotion recognition together.
~ Installation example 2. Utilization at HR (power harassment prevention) ~
~ Installation example 3. Excavation of potential needs ~
Our emotion recognition system has 2 type as [Real time determine] which visualizes emotions and [Batch processing determination] which analyze data later. Here we introduce appropriate use scenes (installation examples) of both types.
Decline of cancellation rate through claim visualization
Customer satisfaction improvement through following up hidden stress
Robot / AI solutions
Enables conversations / responses with considerable emotions
Interviews and internal meetings
Useful to power harassment which is difficult to aware
Personnel appraisal and assessment
Usable of emotions as data materials
Customer needs (potential needs) excavation
Link to potential needs by emotion analysis
The engine specification with energy saving memory / light operation enables the analysis speed of voice recognition process with 10 seconds for 5 sound sources under PC specification of Intel Core i5 / 8GB memory. Real time processing can return the analysis results in about several hundreds mm/s.
※The processing time may differ depending on the processing terminal specification and sound source sampling rate.
Unlike the language reference type used for conventional emotion recognition, our type using Panasonic patent technology is simple and light because it does not require training data in advance. With this, it is possible to perform from the point of installation.
Using the simple algorithm of emotion determination analyzing the voice features, and this enables to use low level specification terminal and Linux environment instead of using special expensive terminals.
Our Voice Emotion Recognition SDK is provided as 「Dynamic Link Library」. This application uses C++ interface, and can input voice data and acquire recognition results. (available of a few other language binding)
We provide SDK format of emotion recognition, so this can be used together with voice recognition engine, and face/expression emotion. By combing with other advanced technologies it can be used as one of AI solutions to acquire higher level and accurate results.
In the specification of our Voice Emotion Recognition SDK, the emotion values are returned from voice data. Specifically, it can acquire the 「rough value)」 got through the emotion strength analysis using the straining condition of the voice codes and the 「soft value)」 got through voice softness analysis, and show each of them by 0〜1 values. the emotion determination can be done by using these analysis results.
<Examples of emotion determination>
◆ Joy : rough value >= 0.5 I soft value >= 0.5
◆ Anger : rough value >= 0.5 I soft value < 0.5
◆ Normal : rough value < 0.5
One of our Voice Emotion Recognition system features is recognition speed, easy of installation/flexibility of customization.
Regarding to the recognition, because its algorithm determines emotions by voice rough values and voice soft values, individual difference occurs hardly and secure the recognition accuracy regardless of languages, dialects.
~ Comparisons of the voice emotion recognition systems ~
|Our emotion recognition SDK||Company A||Company S||Company L|
|Recognition algorithm||Logic rule structure||Logic rule structure||Logic rule structure||Database reference|
|real time recognition speed||several hundred ms||Within 1 second||-||-|
|Recognition emotion type||Anger / Joy / Normal|
|Anger / Joy|
Sorrow / Excitement / Normal
Ups & downs of the mood
|Anger / Sorrow
Excitement / Happiness
|Voice format||Uncompressed PCM||WAV format||WAV format||G711.1 uncompressed|
|Merits||Understand emotional components|
Possible to install into various environments
|Huge research results|
Installation result to Pepper
|Medical business holding|
Possible to determine mood
|Quantification of thoughts
|Other features||Using Panasonic patent technologies||Softbank subsidiary from 2012||Execute depression diagnosis||Need certain length of voice
Mixture of determination results of emotion / thoughts
The installation cost is defer depending on hot to use, equipment's, customize volume.
We will estimate the cost/fee based on hearings. Please contact us.
Our [Emotion Recognition Technology] was installed into [UD talk®]which is the communication support application to materialize visualization of conversations.