Speech-to-Text & Text-to-Speech GenAI API

I use Deepgram  for audio transcriptions and speech recognition. I am working on a feedback survey app where users provide verbal feedback that Deepgram  transcribes into text. 

We receive the results and implement features like punctuation and Smart Format.

Deepgram has significantly improved our transcription process in terms of speed and accuracy. It has allowed us to efficiently convert verbal feedback into text, enabling quicker analysis and implementation of new features. 

Integrating Deepgram has streamlined our workflow, enhancing productivity and delivering more accurate transcription results.

We previously used IBM Watson, which was slow and had limitations in accurately transcribing words. After evaluating OpenAI's Whisper model, we discovered Deepgram, which incorporates Whisper and adds the powerful Nova model.

Deepgram's latency is impressively low, around 0.5 to 1 second, making it a superior choice.

Live transcription could be improved. Sometimes, Deepgram's WebSocket is disposed of due to redundancy issues. Enhanced stability in live transcription would be beneficial.

I have been using Deepgram for one and a half years.

Initially, we encountered some stability issues, but Deepgram has since improved its architecture. With the addition of hooks for status updates, the accuracy has improved to approximately 90 to 95%, which is better than other models we've tested.

It's scalable. Our platform handles 50 to 60 users simultaneously without compromising accuracy. For instance, a 20-minute audio file was transcribed within a second, demonstrating its ability to handle large volumes of audio data effectively.

My experience with customer service and support has been positive. They are responsive and helpful, and they provide timely resolutions to any issues.

Positive

We previously used IBM Watson, but it didn't deliver appropriate results. We searched for alternatives and found OpenAI's Whisper model, which was initially slow. After thorough analysis, we discovered Deepgram. It proved to be superior, leading to our decision to migrate. We used a detailed spreadsheet to compare various models before making the switch.

Thanks to clear documentation, the initial setup was very easy. If you have prerequisite knowledge of the programming language you're using, it’s straightforward to follow the documentation and implement it into your system. When I started, I closely followed the documentation, which made the process very manageable.

Deployment model: We last deployed it on the Google Cloud  Platform (GCP).

The implementation was done in-house.

Our ROI has increased due to enhanced transcription accuracy and speed, leading to more efficient workflows and better user satisfaction.

The pricing is moderate. While live transcription may incur some charges when the connection is open, they become minimal over time. So, it's a balanced option—neither cheap nor overly expensive.

Yes, besides IBM Watson, we evaluated OpenAI's Whisper model.

Deepgram is highly recommended. Users don’t need to do anything special before using it, as the documentation is comprehensive. I am a Node.js developer and have used Deepgram packages for Node.js. Understanding your programming language is key, whether it's Node.js, Python, or others.

AI Features:

I have integrated various AI models into our application. Deepgram's sentiment analysis feature allows us to create graphs and analyses to determine if words are positive, negative, or neutral. This helps us summarize feedback and derive actionable insights.

My ratings:

I would rate it an eight out of ten. The live transcription feature needs improvement as the WebSocket sometimes gives errors or breaks down during live streams.

We use the solution for TTS (Text-to-Speech) and STT (Speech-to-Text) purposes.

The solution's Speech-to-Text conversion feature is really awesome.

Deepgram is currently restricted to only the English variants, but it should include other languages, such as German or French.

I have been using Deepgram for five to six months.

Deepgram is a stable solution.

The Deepgram cloud can handle large volumes of audio data. Around three to four people use the solution in our organization.

The solution’s initial setup is easy.

Deepgram is a cheap solution. We can create an account for $200, which we can initially use for the Deepgram services.

I have used Deepgram with Twilio for the calling system. I would recommend Deepgram to users who want to use it for speech-to-text purposes.

Overall, I rate the solution an eight out of ten.

I am the software development team lead in my company, and we are creating a speech recognition product based on a few different engines. The company works in the legal process, so my team generates a machine-based transcript and then converts it to a readable format. It saves time for the person using the machine-based transcript as a starting point.

The solution's most valuable feature is its speed of transcription. It is one of the fastest tools, especially if you compare it to the second fastest solution that you can get, which is 20 times faster. Thus, it is not just a marginally faster product.

In comparison to Deepgram, I would say that the transcript accuracy offered by other products is much higher. In our company, we had five jobs, and each job was generated by two different engines. One of the aforementioned engines was Deepgram, and the transcribers had two versions, and the users didn't know which one was which version. After the aforementioned process was carried out, the users had to choose which version they thought was less work to convert to the final or the perfect legally binding document since the tool was getting paid per page. Deepgram had a clear incentive to do some little work as fast as possible and to get a particular amount of money. In our company, we didn't have a huge sample size because we are a small company. We did five jobs with five different transcribers, and each job had two versions. A blind test was done, and we found that the other tools were marginally more accurate than Deepgram.

I would like it to be more accurate. I can get maintainability and faster transcripts with the perfect features with an improved tool.

I have been using Deepgram for a year. I use the solution's latest version. I am an end-user of the tool.

I don't use all of the features in the tool because Deepgram adds new features every few months. The features that I have been using in the tool have been very stable. I have never had any issues with the tool and it has never crashed. In our company, we just update to the latest version, and see if there are no issues until the next update.

I use the tool as a software developer and transcriber. Though I don't know the number of transcribers that work in our company , I can say that it runs into hundreds.

I contacted the solution's technical support for help, and I got a nice, decent service and had no complaints at all. At the moment, whenever you want to update Deepgram by yourself, it is a very easy process. You just get the version of the tool and pull it from Docker. If you want to update the model, you have to contact the support anyway. I haven't contacted the support team very often. I might have contacted the product's support team three times just to update the new versions of the model. I got decent support from the tool's support team.

My company still uses multiple products in parallel to Deepgram. For example, in transcript-related business, some clients want to get the transcript super fast or in a few hours. You need to produce the machine transcript in a few minutes and give it to the transcriber, who should start working immediately on it. In such cases, there is a huge improvement when one uses Deepgram, which is a major advantage.

If I consider and compare the other engines I have used with Deepgram, I would say that the ease of installation is one of the strong points of the product. Compared to all other engines, the installation of Deepgram has been simpler and far more stable. It just gets updated, and it runs properly. The good thing is that the tool is modular. In the tool, the modules and the engine itself are separate objects. If you want to update only the module, then there is no need to redeploy anything since most other engines that we have in our company are on an on-premises model.

The solution is deployed on on-premises and cloud environments. It is deployed in the private section of our company's cloud. We aren't using the API and prefer to use our own deployment.

When using Deepgram, one needs to pay for the hours or minutes for which the transcription is needed. The more hours you commit to in advance, the cheaper the price. It is slightly cheaper than the other engines we used. You should take into account that you usually pay for cloud resources or even if you are doing just an on-premises deployment, and since it is a fast process, theoretically, you can save twice, like, once on the money you pay for usage and the second time you pay for cloud resources because if you can, like, finish transcribing in a minute instead of an hour, you can know, if your pipeline scales down, you will only pay for that minute.

Whether you should use the tool or not depends on your use cases and the main use cases where it is used. Based on the engines and factors like maintainability or how easy it is to maintain, and if you consider them to be your priorities, you definitely go with Deepgram.

Speaking about how Deepgram handles large volumes of audio data without compromising inaccuracy, it is always a trade-off. As I said, Deepgram is not as accurate as the other engines we're using, but the difference is marginal, and if speed is more important to you, you should go with Deepgram. I think if the accuracy of the transcript is far more important, other engines give better results.

For an end user, the tool offers on-premises deployment, and it also has an API. Using API is super easy since you just log in to the site and create an account, and you can start using it. If you want to deploy it on an on-premises model, you need to have a basic understanding of the cloud and how it works. The tool has a step-by-step guide on its website, which is very nice, and I still use it even because it is super simple and can be easily understood.

We haven't actually used our AI features at the moment because we're not quite sure how we can use them in our company because we are working on the legal transcript, and those have to be developed word by word even if the person who speaks, says certain things incorrectly. We have to maintain 100 percent accuracy. I am not sure how we can apply AI features in the tool, but we are always looking at the AI aspect in our company.

Deepgram offers some clear advantages over other applications. If you want a tool that produces transcripts in a very fast manner, there is nothing that comes even close to being disputed with Deepgram.

I rate the tool an eight or nine out of ten.

We primarily use the solution for transcribing speech to text. We use it to record phone calls and meetings and then transcribe them.

The quality of the product is the most valuable aspect. The recognition of industry-specific terminology phrases and abbreviations is really important for us. We were able to get a good level of industry specificity with Deepgram.

Two things come to mind for improvement. Maybe they have fixed these, or maybe there is something new, and we haven't implemented it yet. 

One improvement could be dual-channel audio. We've had issues in the past where it generates the transcript, and a lot of the text is duplicated. I understand why it would happen. It's an audio file with more than one channel of the same speaker, which is what may cause the duplicated text. That said, it would be great either to have a way for Deepgram to realize that it's basically the same audio on two channels and only transcribe one of them or at least give us a warning that it's happening. We've found workarounds, however, a better solution from Deepgram's side would be great. 

The other issue comes up when some changes are made on their end, and we want to test them. We've had one to two instances where they tell us that we have access, and we try to test something out, and it turns out we don't. When that happens, then they have to fix something on their end. It's not a big deal. We have a Slack channel with them where we can quickly touch base. We let them know, and they will get back to us and fix the access. It's not something we're doing very often. 

We've been using the solution for about a year and a half.

The solution is stable. We've had no real issues. The only issues that come up more in the testing space are when we're looking at a new model that we want to test out. The issue in that case is more about access where we think we have it and then we don't. That's not a stability issue, however. 

The scalability is significant. We have grown significantly with it and I don't see any reason why we can't keep growing with it. It doesn't require many people to have it up and running. The output that we get from Deepgram is used by a large number of people. Day to day, it's really very hands-off. 

Either before we set up the contract with Deepgram, or right after, they set up a Slack channel with us that had a bunch of people from their end - including customer service and their operations team. We had a call or two with them to get started, and we have been able to quickly correspond with them to get answers or send them details via Slack. That's been very helpful. Even now, we can jump onto the channel and send them questions. They send out updates via email as well.

I can't say if having a Slack channel with them would be considered an extra cost or if it is available to any customer or not. Maybe if you have a premium program or custom model, it's definitely included. 

Positive

Previously, we were using human transcription, which was a lot more expensive.

The initial setup isn't any different from any other speech-to-text solution out there. My understanding is that it wouldn't be extremely difficult. 

It was implemented by one of our developers in-house and I was involved as well. The Deepgram team was very supportive. It depends on the use case, however, in our case, it doesn't require a big team to set up or manage. 

We had our in-house development team involved to get it up and running.

The ROI has been excellent. The cost is night and day compared to the cost of human transcription. We're spending maybe a tenth of the cost we would if we were still doing manual transcriptions.

The pricing was very good. Although the competitors also would have saved us a lot of money, we were mainly looking for the right level of quality of the transcript. 

We compared several other automated transcription services similar to Deepgram a little over two years ago. We're constantly testing out others. We find Deepgram to be the highest quality. That was really important to us.

Amazon AWS has some sort of solution that we tested out. I don't remember the names of the others. We might have tested something from Google and some other lesser-known options.

Being able to access it and send recordings was the same level of work for our development team. So it came to the quality and the ability to build a custom model. We gave them some data and some audio file samples that they used in order to come up with a custom model that would perform even better for us. 

We're a Deepgram customer. 

We use Deepgram for speech-to-text. There are different models you can use in order to do that. We have a custom model and we have also tested out some of their other models as well. 

We're really happy with it. Technology in this space is constantly changing, and we see what's happening with ChatGPT, for example. Everyone uses different kinds of platforms for transcription. Deepgram also has a lot of other solutions we're not using, and it supports other languages, which is not how we're using it. However, it will likely continue to improve itself and align with all of the advances in the space. Deepgram has been really great for our use case. 

I interact with Deepgram's team and the team internally that has implemented it. I don't have hands-on experience with it myself. I continue to review the quality of the transcripts, which is what we're using it for. That said, my understanding is that the ease of use would depend on your experience and your skill level. It is in line with everything else out there. 

I'd recommend the solution to others. I'd rate it nine out of ten. We evaluated and compared different solutions based on the quality of speech-to-text and implementation. For quality, Deepgram definitely came out above everything else. The implementation also went well. Support has been good - not that we've needed it too often. Compared to what we had before, which wasn't a technological solution, the difference is night and day. The other automated transcription services we've looked at in the past few years just weren't good enough. Deepgram really offers great quality. 

I run Deepgram on my local system to transcribe videos.

The speed of the solution for transcribing videos is good.

I need to transcribe my videos to text chat, but there are some issues when I run Deepgram. The solution does not properly identify the number of speakers. For example, Deepgram only identifies two speakers out of three or four speakers in some videos.

The solution also makes some spelling and English grammar mistakes. Deepgram does not properly identify some specific words in a sentence.

I have been using Deepgram for one to two months.

We haven't faced any breakdowns or bugs with the solution.

My team consists of two members who use Deepgram.

The solution’s technical support is average. I talked to the technical support team regarding an issue where the solution couldn't identify the exact number of speakers. The support team asked me to use certain parameters, but the results were inaccurate. I used all the parameters suggested by the support team, but the speakers were still not identified clearly. However, other services could properly identify the speakers of the videos.

The solution’s pricing is cheap.

I chose to use Deepgram after researching it on Google and finding some good feedback that the solution had good APIs. It's easy for a new user to learn to use the solution.

I would not recommend Deepgram to other users because it does not properly identify video communication. If you compare it with the other APIs, you can easily find that they do not properly identify some words, exclamatory signs, full stops, etc. These are small mistakes the tool is not properly identifying. Also, the solution does not properly identify the speakers. Users should check other APIs before choosing Deepgram.

Overall, I rate the solution a six out of ten.