Deepgram is really a developer-friendly solution and is extremely fast if you compare it against other vendors like OpenAI Whisper and Speechmatics. I mention friendly because I never had to reach out to support to get the integration working, is pretty easy, well-documented, and self-explanatory. Something that caught my attention about making the switch is the ability to train domain/industry specific vocabulary; this is very useful when you need to focus your solutions in the industry, reducing friction for the final user. Review collected by and hosted on G2.com.

There is a big data privacy trade-off. If you leave the Model Improvement Program that uses the audio data to train their models, you lose the 50% rate discount, this is bad for FinTech startups, where money is a valuable asset that you need to take care of, but as a highly regulated industry, you can't share your data easily; the previous thing combined with the modular pricing means that the cost add up quickly when you need advanced features. Review collected by and hosted on G2.com.

What I love most about Deepgram is the accuracy and speed with which it transcribes. I've tried other solutions and always missed some details, but with Deepgram, I feel the result is much more reliable. Additionally, the integration is simple, and the platform is very intuitive, allowing me to focus on my work without wasting time configuring things. I also highlight the flexibility of its models, which adapt well to different accents and contexts. In summary, it gives me peace of mind knowing that I can rely on a tool that truly understands what is being said. Review collected by and hosted on G2.com.

What I like least about Deepgram is that, although the accuracy is very good, it can sometimes be a bit difficult to adjust the models for very specific cases or when there is technical jargon. I would also like the documentation to be clearer in some points, because at first, I had to spend extra time to fully understand how to take advantage of all the features. These are not serious problems, but details that could be polished to improve the experience and make it even smoother. Review collected by and hosted on G2.com.

i mostly used the STT (speech to text) with nova 3 and the best part is that it supports words recognization and speaker diarization and it also have a very good price offering comparatively

easy to integrate and change the parametes

quick support whenever required

and the performace it is very much statisfied

overall it is very good platform for STT and TTS

also it forms a user friendly portal to test out the api without putting any efforts in integaration. Review collected by and hosted on G2.com.

Deepgram performs well overall, but I have experienced occasional inaccuracies with speaker diarization. Sometimes speakers are not distinguished correctly, or the labels overlap, which can affect the accuracy of multi-speaker transcriptions. Review collected by and hosted on G2.com.

The new Voice Agent API is the feature I love the most. I have been using Deepgram for the low latency STT but now with this API I can involve LLMs directly into the context and I don't have to call another API just for using the LLM. Love this new integration. Excited to see what other build with it. Review collected by and hosted on G2.com.

Deepgram has grown a lot since I last saw it. It's heading in the right direction for growth. The only concern is if someone has already signed up to Deepgram now we can't get that $200 for using the Voice Agent API and Audio Intelligence. Review collected by and hosted on G2.com.

Deepgram has beena godsent to us! We run Deepgram as the transcription backbone for a language-learning app that imports tens of thousands of short-form videos a month, and the thing we value most is that a single prerecorded call gives us everything downstream needs: word-level timings, diarization, smart punctuation and utterances, which is what our karaoke captions, speaker-aware segmentation, auto-pause and word-tap glossing are all built on. The language=multi code-switching mode on Nova-3 is genuinely differentiated for us, since per-word language tags let us detect and handle a Spanish creator dropping into English mid-sentence, and no other provider we tested returns that. Language coverage across Nova-2 and Nova-3, including dedicated regional models like Swiss German, effectively determines which languages we can ship at all, and tokenization on spaceless scripts such as Chinese, Japanese and Thai is consistent enough that our whole difficulty and CEFR scoring model is calibrated on it. Operationally it is easy to live with: we can pass either a buffer or a remote URL, so re-transcribing from stored audio costs us nothing extra, and unrecognized model/language combinations fail cleanly enough that a simple retry handles them. Review collected by and hosted on G2.com.

We have a lot of learner demand for Esperanto, but sadly it's not available yet. :) Review collected by and hosted on G2.com.

The accuracy of transcriptions is flawless with Nova -3 Model.Ultra Low latency with almost realtime audio transcription makes it a best choice.Also, The smart formatting and multi-lingual language changes make it best for production ready deployment. Lastly the flawless Voice agent API manages the real time connection loop.It has a intuitive UI for easy usage. Review collected by and hosted on G2.com.

The cost can be a factor as the scale goes up the pricing rise.It does have technical complexity which requires to do spend time on learning execution. Review collected by and hosted on G2.com.

I work as a Dietician and Nutritionist, manage a homeopathy clinic, create health education content for social media, and also prepare educational material for patients and students. What I like most about Deepgram is that it helps convert spoken content into text quickly and in a very organized way. I tested it with a health awareness audio recording, and the transcript was generated fast and was easy to review. What makes it useful for my work is that it offers more than basic transcription. Features like speech-to-text, smart formatting, speaker recognition options, summarization, topic detection, and audio file transcription help me work with recordings more efficiently. When I create health education content, prepare awareness campaigns, review educational recordings, conduct online teaching sessions, or organize information for patients, these features help me identify important points much faster.

The performance has been good during my use. Transcripts were generated quickly, the text was easy to read, and the formatting required very little cleanup. I also found the summaries and detected topics useful because they provide a quick overview of longer recordings. This helps save time when preparing patient education material, nutrition guidance notes, health awareness content, presentation outlines, and social media posts. Another thing I like is that it reduces the effort needed to review longer audio files. Instead of repeatedly listening to recordings, I can work directly with the transcript, summary, and detected topics to organize information more efficiently. This is especially helpful when handling educational content, clinic related recordings, and health discussions that need to be converted into written material. I do not use it every day, but whenever I need to work with recorded content, Deepgram helps make the process faster, more organized, and easier to manage. Overall, Deepgram helps save time, keeps recorded information organized, and makes it easier to work with audio-based content in a more structured way. Review collected by and hosted on G2.com.

The transcription quality is generally very good, but with recordings that contain strong accents, unclear speech, or multiple people speaking at the same time, I sometimes need to do a small amount of manual review. It does not happen often, but it can occur with real world audio. Review collected by and hosted on G2.com.

I really like that Deepgram powers all of the transcripts for our podcasts with great accuracy, especially when dealing with technical language and specific terms. I appreciate how it allows us to upload specific terminology, making our transcripts much more accurate than other services. It's impressive how Deepgram lets us automate the entire process, from downloading podcasts to getting back transcripts that need minimal editing. What I love most is that Deepgram works seamlessly in the background without needing manual intervention, making our work faster with less effort. Review collected by and hosted on G2.com.

I think maybe the capacity to share more terms would be nice. And, I'm not sure that it gives us everything we want in terms of speaker identification. I think that maybe could be improved. We just haven't been able to find a programming process that will give us back all the identified speakers as accurately as we would like. I'm not sure if I'm able to upload samples, for example, of each speaker. But, something like that would be useful if I'm not. Review collected by and hosted on G2.com.

What I like best about Deepgram is the real-time transcription speed and accuracy — the Nova-2 model processes audio almost instantaneously with remarkably low word error rates even on technical jargon and domain-specific vocabulary, which was a persistent failure point with other transcription services we evaluated. Review collected by and hosted on G2.com.

One consistent frustration is the documentation gaps around edge cases - while the core API docs are clean and well-structured, when you move into advanced configurations like combining smart formatting with custom models or fine-tuning diarization sensitivity, the documentation becomes sparse and I've had to rely heavily on community forums and trial-and-error to find answers. Review collected by and hosted on G2.com.

I use Deepgram for both speech-to-text (STT) and text-to-speech (TTS) in AI applications and workflow automation. It provides accurate transcription across multiple languages and produces natural-sounding speech with low latency, making it suitable for real-time and batch processing. The biggest advantage has been its language support and output quality. The transcription accuracy is consistently high, including for accents and multilingual content, and the TTS voices sound clear and natural. The API is straightforward to integrate, the documentation is well organized, and the response times are fast. Overall, Deepgram has been a reliable solution for building voice-enabled applications and automating audio processing tasks. The initial setup is also very easy. Review collected by and hosted on G2.com.

Deepgram performs very well overall, though there are a few areas that could be improved. I would like to see broader language and dialect support, particularly for less commonly used languages and regional accents. The TTS voice selection could be expanded with more natural and expressive voices across additional languages. Pricing for high-volume usage could also be more competitive, especially for developers building large-scale AI applications. More built-in features for speaker diarization, custom pronunciation, and voice customisation would make the platform even more versatile. Review collected by and hosted on G2.com.