Embracing the real-time revolution

We live in a world of instant.

Gone are the days of slow, plodding dial up internet with small files taking hours to download, or packages taking ages to arrive, or waiting weeks and weeks for episodes of TV shows to air. Everything now is always on, instant, simultaneous. It's real-time.

Instant streams, notifications, messages, insight, updates, workflows, understanding. The world of business and software has been accelerating for decades and shows no signs of putting its foot on the brakes.

A new world of possibilities

Customer expectations have risen in parallel. Rare is the occasion that we as users are happy to wait for slow, especially in the world of software.

This is also not just impatience. Some things just can’t wait.

Captions on live television are useless if it takes them minutes to appear. Information used to help a customer on a call is far less effective if it is sent days after the conversation took place.

Real-time, live, instant, low latency. Whatever you call it, the transcription and use of audio ‘in the moment’ is here, and has now reached a point where the accuracy is so reliable, that a whole new world of possibility has opened up.

More options for interactivity, more instant insight, broader international audiences, more engaging content, workflows powered by voice, and so much more.

Real-time ASR and Speech Intelligence has arrived.

Limitations of batch and file by design

The traditional model of Automatic Speech Recognition has always been to record a media file (either audio or video) then send it away for transcription.

Depending on the length of the media, this may have taken seconds, or even several minutes. This way of transcribing limits how quickly you can start getting value from the transcript.

For some use cases, this has been entirely satisfactory.

File, or batch transcription still has many benefits, and will be great for many use cases. Compliance related tasks, for example, where recordings of calls and audio need to be stored and audited, do not have a need to be done instantaneously. Calls used for training are only done so after the fact. 

Similarly if subtitling or captioning recorded media – if there is any trade off at all between accuracy and speed, in this case, accuracy will always win. In instances where cost is also the priority, users will prioritize that over the need for either accuracy or speed.

Trade-offs are always made, and made for good reasons. 

But, what happens when time is no longer a barrier?

The power of now

In the above example, the fastest a transcript could be used was the 1 hour call plus the time it took to return the transcript. We might generalize this to:

Time to value = length of media + time to return transcript

This simple equation contains some big unknowns and variables.

This might be good for media only a few seconds or minutes long, but take a media monitoring company, who is recording the Super Bowl to check for mentions of a brand. The Super Bowl can be over four hours long. But media intelligence requires agile responses to mentions and sentiment. This is just too long to wait. 

So, in the world of Speechmatics’ real-time, how long would you have to wait to be able to do something with a transcript?

Two seconds.

No variables or unknowns here.

Transcription that you can trust 🤝

Incredibly, machine learning powered transcription from Speechmatics has reached a stage where you can now get transcription that you can trust within just two seconds (with the option to bake in delays of up to 10 seconds if you want maximum accuracy). For those who are looking for faster transcription, we can return words in less than 1 second.

The six benefits of this are HUGE: 

1) Value, without the wait

With file transcription, you have to wait for the entire conversation or media file to be finished. With real-time, you can start using the voice data to provide value to the user immediately. You can provide instant insight, assistance, analytics, guidance, and understanding.

2) Far more than just English

With Speechmatics, you’re not limited to real-time transcription in English. Every single language we offer, we also offer real-time transcription for, without compromising on accuracy. So no matter the market you’re in, speech technology can be used.

3) Transform customer experiences

Users can fix issues faster, provide better customer support, and your products become more inclusive. Your end users have access to new features, without waiting for files to be uploaded, transcribed, and returned. This allows them to be more agile, faster moving, provide unbeatable customer experiences.

4) Instant - without losing accuracy

Historically there has been a huge trade off between accuracy and speed. This no longer exists. Speechmatics has accuracy comparable with file or batch transcripts, so any workflow or feature built on top of real-time transcription will still be based on rock solid foundations.

5) Increased engagement

For any live broadcast or media, accurate real-time transcription can maximize your audience engagement, including those with hearing impairments and those wanting to watch in an environment prohibitive of listening to the audio.

6) Speech Intelligence... in real-time

Providing a transcript in real-time is only the beginning now that the AI era has truly arrived. With breakthroughs in LLMs and our speech capabilities, you can stack insight and value on top, without the wait. This includes Sentiments, Topics, Summaries and much more.

A new world of use cases

Real-time unlocks a world of new use cases, features, and value that can be delivered to people using products and services.

• Instant workflows powered by voice Processes and workflows based on words or phrases that can kick start instantaneously – be that brand mentions, requests for account upgrades/downgrades, company specific workflows or actions mentioned during a meeting that do not have to wait until later.

• Really smart Interactive Voice Response Voice bots and call routing that can understand more complex utterances and respond quickly and effectively to queries, drastically improving customer experiences in a scalable and cost effective way.

• Agent assist Knowledge base articles and AI-driven advice to customer service and sales agents, giving people empowerment and technology to help them be incredible at their jobs.

• Live social content Accurate captions for live streamed content across social media platforms, so that even those who cannot crank up the volume can be involved and engaged.

• International audiences for everything Live transcription and translation of audio in multiple languages to increase the global reach of content, and ensure audiences can tune in without a delay.

• Voice commands in automotive systems ASR enables drivers to control in-car infotainment systems, navigation, and other features with voice commands, helping reduce distractions while driving.

• Voice search for e-Commerce Real-time ASR can enhance the shopping experience by allowing customers to search for products or services using voice commands in online retail applications.

• Healthcare In medical settings, ASR can transcribe doctor-patient interactions, record medical notes, and help with real-time documentation, improving the efficiency of healthcare providers. 

For every product or service that handles media or audio, there will be dozens of valuable features that will benefit from the power of now. 

This is speech technology fit for the modern world of instant, always on, fast-paced, move it or lose it. This allows your customers to react as quickly as they need to, and stay ahead. The waiting for batch files to be returned is over.

Building Speech Intelligence with live speech data

The above examples are incredible in their own right, but we’ve not even scratched the service with them. Large Language Models and AI had a revolutionary 2023, with these services hitting the mainstream in a big way. Their versatility and intuitive usefulness have made them a technology that every company is looking to harness. 

They can also be used with live audio. 

Last year, we launched several outstanding speech capabilities:

• Translation

• Summaries

• Topics

• Chapters

• Sentiment

These harness the innate abilities of LLMs to provide additional insight and understanding into audio data.

Driving value with speech capabilities

When applied to accurate, real-time transcription, these capabilities can become enormously value driving.

Summaries in real-time can be a time-saver for both meetings and contact center agents, but also allow them to view the key talking points up until the current moment. People are more informed about previous information shared, and can instantly share or submit the notes with others or as part of their process.

Sentiment can drive more responsive and agile media intelligence – for many brands, when launching new campaigns, will want to manage budgets and spend to match the fast-paced nature of media. If a campaign is being received well, where should we double down, or where should we back off?

Speech Intelligence is the pursuit of value using the spoken word. With communication being free-flowing and fast paced, often this cannot wait until after the fact. Real-time in combination with the latest breakthrough in AI, represent something special, and we know that our customers and users will build innovative features and products we haven’t even thought of yet using this new area of innovation.

A step towards seamless interaction with technology

Real-time also represents something even bigger and longer term. Even though products like ChatGPT can now be interacted with using voice, these interactions are still a way off being truly seamless. They remain a call-and-response, or question-and-answer style dialogue. 

This is not how we naturally communicate. 

Speechmatics envision a future where interaction with technology is not about excellent prompt engineering. Instead, it’s where technology works hard to understand our intentions and communicate with us in a way that feels natural, not stilted and formal.

Tackling challenges and seizing real-time ASR excellence

In order to get there, we have some big challenges to overcome. But we are committed to this journey, and will update you as we go.

In the meantime, we are excited by the inflexion point that accurate real-time ASR represents, and urge everyone in the business of voice to get stuck in.

You can evaluate Speechmatics’ real-time capabilities via our portal, today.

So, what are you waiting for?