I ate the world's hottest pepper to get you to respond to this email. Your move.
Transcription provided by Deepgram
This is the Trinidad Scorpion. It's two million Scoville units. If I eat this, You've got to reply to this email. Right? Okay, let's do this.
I'm Scott, CEO of Deepgram. We build world class deep learning based speech recognition engines, and we work with companies that are frustrated with the low quality and high cost of Google, Amazon, IBM Watson. We know you've been working with those companies, but it's painful. I need more ice cream.
Everything we do is based on end to end deep learning on GPUs. It's really, really hot. NVIDIA is a very tight knit partner with us. We take those models, our deep learning models, and put them on GPUs. You get amazing results. It's really fast. It's really, really accurate. And it's extremely easy to turn around when you want an update. It really burns a lot. It's getting very numb. But we're we're the best.
Compare us to Google, compare us to IBM Watson, come do an evaluation with us. It hurt. We'll take some of your data will show you how much better we are. You can have world class scientists making a model for you. So loud. So loud. But Mhmm. It tastes so good. It's Deepgram ice cream, it susses your pan very burning tongue.
In case you're curious, here's how we extracted the transcription:
Step 1
First, we needed to download the video locally. There are several ways to do this but we used a handy tool called, "youtube-dl". Install info here.
youtube-dl https://fast.wistia.net/embed/iframe/275gfw0348Step 2
That gives us dg-chili-pain-highres-275gfw0348.mp4 as a local video file.
Step 3
Next, we need to extract just the audio from the .mp4. We use ffmpeg. Install info here.
ffmpeg -i dg-chili-pain-highres-275gfw0348.mp4 -ac 2 -f wav dg-output-audio.wavStep 4
Okay! Lots of work just to extract the audio file but now we're ready to generate the transcription via the Deepgram API. You'll need a Deepgram account and an API key to complete this final step.
This is what generates the json transcription output. (you can remove the | jq if you don't have that package. It makes the json return value formatted and pretty. Download jq here.
curl \
-X POST \
"https://api.deepgram.com/v1/listen?punctuate=true&model=nova&language=en-US" \
-H "Authorization: Token YOUR_SECRET" \
--data-binary @dg-output-audio.wav | jqResults
Here's what you should end up with: (removed the "words" node to save space).
{
"metadata" : {
"channels" : 1,
"created" : "2023-06-17T00:31:47.090Z",
"duration" : 106.58138,
"model_info" : {
"3312b52b-d08e-4072-bfff-124c410b770d" : {
"arch" : "nova",
"name" : "general-nova",
"version" : "2023-03-13.31000"
}
},
"models" : [
"3312b52b-d08e-4072-bfff-124c410b770d"
],
"request_id" : "19243606-d125-4f49-ada0-c6ff6e3ec0fb",
"sha256" : "0be9dd0945346c8a5b786152903ea0354734863487f91bcb877861375a0ac9a1",
"transaction_key" : "deprecated"
},
"results" : {
"channels" : [
{
"alternatives" : [
{
"confidence" : 0.9946289,
"transcript" : "This is the Trinidad Scorpion. It's two million Scoville units. If I eat this, You've got to reply to this email. Right? Okay, let's do this. I'm Scott, CEO of Deepgram. We build world class deep learning based speech recognition engines, and we work with companies that are frustrated with the low quality and high cost of Google, Amazon, IBM Watson. We know you've been working with those companies, but it's painful. I need more ice cream. Everything we do is based on end to end deep learning on GPUs. It's really, really hot. NVIDIA is a very tight knit partner with us. We take those models, our deep learning models, and put them on GPUs. You get amazing results. It's really fast. It's really, really accurate. And it's extremely easy to turn around when you want an update. It really burns a lot. It's getting very numb. But we're we're the best. Compare us to Google, compare us to IBM Watson, come do an evaluation with us. It hurt. We'll take some of your data will show you how much better we are. You can have world class scientists making a model for you. So loud. So loud. But Mhmm. It tastes so good. It's Deepgram ice cream, it susses your pan very burning tongue.",
}
]
}
]
}
}There you have it! After we had the audio file from the video, the transcription only took a few seconds. That shows you how fast Deepgram performs.