Artificial Intelligence and Machine Learning have opened opportunities for automatic speech recognition with many potential applications, amongst others for customer experience improvements and automation in call centers. While the likes of Siri and Alexa are already quite advanced in understanding the most widespread languages, viable solutions for smaller languages are lagging behind.
Swisscom’s AI and Machine Learning Team is currently tackling the challenge of understanding Swiss-German and other less spoken languages.
For Swiss-accented French or Belgian-accented Dutch (Flemish), that are closer to standard languages, Swisscom and Proximus are in search of the most advanced innovators:
Call for Innovation #3:
Automatic Speech Recognition and Sentiment Analysis of Swiss French and Belgian Flemish
Leading Swiss telecommunications provider Swisscom, together with Proximus from Belgium, reach out to international software vendors and startups. In this Call for Innovation, we aim to find the most advanced innovators in automatic speech recognition (ASR) and sentiment analysis.
If you are a startup or innovator that has successfully developed automatic speech recognition solutions and can deliver high accuracy in analyzing conversations in Swiss-accented French and/or Flemish, we are inviting you to apply until April 11, 2018.
All suitable applicants will get access to training data to fine-tune existing models until May 28 2018, when we will test the solutions in a live challenge with similar voice data.
The winners will be selected based on accuracy and latency of the analysis as well as further evaluation by our experts. They have the chance to start a paid Proof of Concept to further verify the capabilities and integration in our existing environment.
The Challenges in detail
Based on possible use cases for automatic speech recognition and analysis, we have defined one specific and one open challenge. Applicants are welcome to participate in both or only individual challenges:
1) Call Center Challenge
Call centers will always remain an important touchpoint in the telecommunication industry. Yet, artificial intelligence allows automation of routine work and enables our agents to focus on complex or emotional tasks in personal interactions. Possible use cases may range from voice-based identification of customer intent to live support and automated call processing. Thereby, we aim to improve both operational excellence as well as customer satisfaction.
To leverage the full potential for call center automation and build up relevant capabilities in Swiss-accented French and Belgian-accented Dutch (Flemish), the challenge consists of two independent parts:
(a) Automatic Speech Recognition of a call center conversation
(b) Sentiment analysis throughout the call center conversation
Training data will consist of about 50 hours of annotated mono voice recordings of:
- actual call center conversations among a Swisscom agent and customer in Swiss-accented French;
- Public content (e.g. interviews, lessons, news reports and phone dialogues) and a sample of actual call center conversations in Flemish.
What we are looking for? For ASR, we expect accurate (near) real-time transcriptions and primarily target innovators that already have a trained model for (Swiss-accented) French and/or Flemish. Sentiment analysis will be evaluated based on the level of Detail.
2) Open Challenge
Beyond call center automation, voice identification, and voice control, probably many more use cases for ASR and sentiment analysis exist. In the open challenge, we invite startups and innovators to pitch innovative solutions in any language. More specifically, we want to explore the potential of ASR- and sentiment-based concepts that have already achieved some traction in test markets (i.e. paying/very interested business customers or tested high value for private customers).
No specific training data will be provided for the open challenge but you are free to use the recordings from the other two challenges.
What we are looking for? Out-of-the-box ideas and new approaches with a proof of concept or business case to use the full potential of ASR, sentiment analysis, and the like. This challenge is language-independent.
Why should you participate? All finalists will be invited to pitch their business case and propose a proof-of-concept. Our experts will carefully evaluate each suggestion based on the business potential or possibility to enhance customer and/or employee experience. Winners get the unique opportunity to realize a paid proof-of-concept with our development team.
What do you have to hand in? We expect a pitch of your business idea. Please provide us a description of it and any additional documentation. You will see the respective fields in the application form. For the evaluation we will set up a Skype Call to hear you pitch after May 28, 2018.
We invite applications from encouraging startups, small innovators, and more established companies from all around the world that can provide accurate speech-to-text transcripts and analysis.
To participate, you…
- support on-premise deployment of your software,
- are ready to test your solution in a (paid) proof-of-concept later on,
- ideally already have reference customers in the domain of call centers and/or telecommunication,
- adhere to the data protection regulations of the EU and Switzerland.
- and are willing to accept the data policies of this Call for Innovation (all participants need to sign NDAs in order to receive the training data).
Automatic Speech Recognition
- Word Error Rate (WER)
- capacity to deal with noise
- Changes in word rate
- identification of tone, intonation and pitch changes
- satisfaction/sentiment metric and its development throughout the conversation
- Detail of analysis
- Business potential of the idea
Process & Application
Application is open until April 11, 2018.
By clicking on the application button below, you will be redirected to a questionnaire. If you have any questions, please do not hesitate to contact us.
For the Open Challenge the application remains open until May 28, 2018.
The Call for Innovation consists of the following six phases:
- Application: Innovators can apply until April 11, 2018 to participate in the Call for Innovation. For the Open Challenge the application remains open until May 28, 2018.
- Training: After an eligibility check, all selected innovators will receive annotated training data on April 23, 2018. During the next four weeks, they have the chance to fine-tune their models and prepare for the challenge.
- Pre-Challenge: On May 28, 2018, we will test the accuracy of all solutions. Participants receive test data (similar to the training data but without transcripts) and we expect to get back analyses in (near) real-time (specifics to be defined).
- Evaluation: The expert jury will evaluate the accuracy of the transcripts and sentiment analysis by June 18, 2018 and nominate the finalists that are invited to the live challenge
- Live Challenge: Inform participants on June 25 - 29, 2018 real-time challenges
- Announcement of winners: The expert jury will evaluate the accuracy of the transcripts and sentiment analysis by July 5, 2018 and nominate the winner(s) of the challenge.
Swisscom is Switzerland’s leading telecom provider and one of its foremost IT companies,headquartered in Ittigen, close to the capital city, Bern. When it comes to serving customer needs, Swisscom brooks no compromise; it focuses on service and quality and invests massively in the networks of the future. The more than 20,000 employees generated sales of CHF 5.69 billion in the first half of 2017.
In the domain of artificial intelligence, Swisscom offers a range of enablers to business customers and has already introduced a variety of intelligent voice-based services for its private customers, like VoicePrint or VoiceSearch on Swisscom TV 2.0.
Proximus Group is a telecommunications & ICT company operating in the Belgian and international markets, servicing residential, enterprise and public customers. Proximus’ ambition is to become a digital service provider, connecting everyone and everything so people live better and work smarter. Through its best-quality integrated fixed and mobile networks, Proximus provides access anywhere and anytime to digital services and easy-to-use solutions, as well as to a broad offering of multimedia content. Proximus transforms technologies like the Internet of Things (IoT), Big Data, Cloud and Security into solutions with a positive impact on people and society. With 13,633 employees, all committed to offering customers a superior experience, the Proximus Group generated an underlying turnover of EUR 5,871 million end-2016.
Proximus (Euronext Brussels: PROX) is also active in Luxembourg through its affiliates Telindus Luxembourg and Tango and in the Netherlands through Telindus Netherlands. Its subsidiary BICS offers best-in-class international wholesale solutions for voice and mobile data service providers worldwide.