Skip to content
4 min read

A Day in the Life: Navigating the World of Voice-to-Text Technologies

Today, I dove headfirst into voice-to-text technologies, driven by curiosity and a dash of professional necessity.

My toolkit for the day? An eclectic mix of solutions: Deepgram Speech-to-text, OpenAI Whisper, AWS Transcribe, Azure Real-time Speech-to-text, Google Speech-to-Text AI, and…. my coffee.

Each promised to transform spoken words into written text - as coffee transforms my sleepy thoughts into coherent ones - but as I quickly discovered, not all tools are created equal.

Welcome to my Morning Routine: Setting the Stage

My morning began with the usual lethargy but a palpable sense of anticipation.

The first order of business (after drinking my coffee) was to navigate the setup process for each of these tools.

OpenAI Whisper was my first contender. Its open-source promised a straightforward setup, and I can say it has lived up to its reputation. Within minutes, I was up and running (no, I don’t have a treadmill, but you know what I mean). However, it soon became apparent that real-time transcription was not in its arsenal—a minor setback but noteworthy for those needing instantaneous results.

Next up was Deepgram, which turned out to be a breeze to integrate. Its ability to deliver real-time transcription was impressive, setting a high bar for its competitors. On the other hand, AWS Transcribe tested my patience with its never-ending manual authentications—a sharp contrast to Deepgram's user-friendly approach.

Google's service faltered at the demo stage, leaving much to be desired, while Azure, though competent, didn't quite make the cut for the in-depth testing phase - no hard feelings here; we just chose to focus on two contenders.

The Midday Experiment: Real-Time Trials

As the day progressed - and after my third coffee - I narrowed my focus to Deepgram and AWS Transcribe. I orchestrated a series of simulated conversations, ranging from the basic to the complex.


Deepgram emerged as the clear frontrunner, its accuracy barely faltering even in the face of ambient noise—a common adversary in real-world applications.

While valiant in its efforts, AWS Transcribe struggled with accuracy, often missing words or misinterpreting phrases, especially when background noise was present.

An interesting observation was AWS Transcribe's tendency to revise its initial transcriptions based on subsequent input—a feature that, while innovative, occasionally frustrated me due to its inaccuracies.

Afternoon Deep Dive: Customization and Integration

You’re familiar with the post-lunch urge to nap? Well, it didn’t stop me from experimenting.

The afternoon was dedicated to pushing these technologies to their limits. I experimented with custom vocabularies, different microphone qualities, and varied speech tempos to see how well these tools could adapt.


Both Deepgram and AWS Transcribe offered the ability to integrate a custom dictionary, a feature I found particularly useful for industry-specific terminology.

Deepgram, however, stood out for its ability to seamlessly adapt to these customizations, consistently delivering accurate transcriptions even under challenging conditions.

Data privacy is a significant concern in my line of work, making the autonomous hosting option offered by Deepgram and OpenAI Whisper particularly appealing. This feature ensures that sensitive data remains within the confines of our network, a crucial consideration for any organization prioritizing data security.

🥖 Do you speak French?

If you didn’t know, HalfSerious is based in Montreal. In Quebec, we speak Canadian French. So, how did the transcription go? Well, it has some room for improvement. It is difficult to interpret words pronounced with a Quebecer accent correctly, and the tests were inconclusive for similar words, like “quel” and “quelque.” Nonetheless, we have a winner, and it’s Deepgram. The transcription was better and faster (compared with Transcribe), so it’s a clear “OUI” for us.


Evening Reflections: Drawing Conclusions

As the day wound down, I reflected on the insights gained from this deep dive into voice-to-text technologies (I also dreamt about the pizza I was going to order, but that’s another subject).

Deepgram and OpenAI Whisper’s ease of setup and accuracy were standout features that distinguished these tools from their peers.

When it came to the cost-quality ratio, Deepgram and Azure offered the best value, providing high-quality transcription services without breaking the bank.

Despite its higher cost, AWS Transcribe did not meet the same standards of accuracy, making it a less appealing option.

The Final Verdict

After a day filled with exploration and experimentation - and too many snacks - my choice was clear: Deepgram, with its real-time service, superior accuracy, and competitive pricing, emerged as the winner.

But this journey was not just about finding the right tool; it was a reminder of the incredible advancements in AI and the potential for these technologies to revolutionize how we interact with digital platforms.

Thanks for following me for the ride. Now I’m off to my pizza. 🍕


Deepgram Speech-to-text, OpenAI Whisper, AWS Transcribe, Azure Real-time Speech-to-text, Google Speech-to-Text AI.

Do you want to receive our new articles in your inbox?