Blog

Late last year, OpenAI announced Whisper, a new speech-to-text language model that is extremely accurate in translating many spoken languages into text. The whisper repository contains instructions for installation and use.

tl;dr:

# Install whisper and its dependencies.
pip3 install git+https://github.com/openai/whisper.git 

# (When needed) Update whisper.
pip3 install --upgrade --no-deps --force-reinstall git+https://github.com/openai/whisper.git

# Make sure ffmpeg is installed.
brew install ffmpeg

# Translate speech into text.
whisper my_audio_file.mp3 --language English

One thing I do quite regularly for my YouTube channel is extract the audio track, convert it to text using an online tool (I used to use Welder until they were bought out by Veed), and then hand-edit the file to fix references to product names, people, etc.

tl;dr: Use an old-fashioned USB 2.0 flash drive, format it FAT32, download the firmware, make sure it's named correctly, and use the motherboard's 'BIOS Flashback' option after powering off the computer.

This past week, I devoted far too much time to the task of updating my brand new motherboard's BIOS.

It started with a combo deal from Micro Center: a ASUS ROG Strix B650E-F Gaming WiFi motherboard, a Ryzen 9 7900x CPU, and a G.Skill Flare X5 Series 32GB DDR5-6000 memory kit, all for $599. Quite a beefy upgrade for the main PC I use to compile code and do random Linux-y tasks.

Recently I recorded an entire video for my YouTube channel using only a Rode Wireless GO II lavaliere mic.

I typically spend the time to set up a second mic source—usually my shotgun mic into a separate recorder—but this time I was feeling lazy. I had never had an issue with the wireless lavs in my basement, and the Rode system includes a built-in recorder in the bodypack transmitter so I have backup audio that has saved my bacon a few times when interference did cause cutouts to the camera input.

But because of that overconfidence, I had to reshoot the entire video (I tried removing the RFI using iZotope RX 10, but there were parts where the interference was still too prominent). Lesson learned: always have the backup audio.

During the reshoot, I still relied on the lav for my primary mic, but it still had the interference, even though I set my phone and iPad into airplane mode, and made sure all WiFi devices within about 20' were powered off!

I made the mistake of putting my Wireless Go II mic transmitter in the same pocket as my iPhone for a recent video recording, and as a result, I had a lot of RF interference in the recorded track.

Thinking I could just use the nice feature of the Wireless Go II's built-in recording, I grabbed the track off the body pack itself—but found that it, too, had the RFI sound, meaning the iPhone's interference made it into the mic circuit itself, not just the wireless mic signal to my camera!

I tried Final Cut Pro's built-in voice isolation, and that helped mute the noise between speech, but during speech it was omnipresent.

I also tried accusonus' denoise plugin (RIP after accusonus was bought out by Meta), and it did better, but left the sound feeling 'watery'.

Radxa's Rock 5 model B is an ARM single board computer that's 3x faster than a Raspberry Pi. And that's just the 8-core CPU—with PCI Express Gen 3 x4 (the Pi has Gen 2 x1), storage is 7x faster! I got over 3 GB/sec with a KIOXIA XG6 NVMe SSD.

It's still half as slow as modern ARM desktops like Apple's M1 mini, or Microsoft's Dev Kit 2023 (see my review here). But it's way faster than a Pi, it comes with 2.5 Gig Ethernet, it has two M.2 slots on board... and, well—it also starts at $150!

...that was the question I asked my Dad, a radio engineer for many decades, who worked at the biggest AM station in St. Louis, KMOX. The station is approaching its centennial in 2025, as are—some YouTube commenters argue—its primary audience!

I recorded that video during my convalescence at my parents' house (I am feeling much better now, thank you!), and my Dad discussed a few reasons why AM radio—at least in the US—is not dead. But it is suffering.

In the video, I pointed out the current dichotomy: