Hello,
I'm trying to learn Swedish from German primarily, so I'd like my lookup language (double-clicking a word) to also be in German, but it's in English.
Screenshot 2025-12-26 at 17.02.58.png
Where does the lookup dictionary call its dictionary from? Even when "deleting" all other dictionaries but DE-SV, I'm getting the same Swedish-English translations.
Vielen Dank / Tack så mycket..
Hey River,
Thanks so much for your work on this project, I have loved using it so far. I've contributed some dictionaries and bought you a coffee too 🙂
A quick question, is it possible to download the word history with the sentence column? Right now it just downloads with things like the url source in the csv while, which isn't that helpful.
Also, is it be possible to modify what is added to the input fields in the anki 'add' window? The extra formatting that is added automatically is making it difficult to make my new cards match my normal formatting.
Currently, when looking up any Chinese word in the popup, the headword and definitions are always in Simplified Chinese, even if the selected word is Traditional and the text is Traditional. I did not find a setting to change this. It would be very helpful to be able to choose between Simplified and Traditional.
9ac7397b-227f-407a-ae7b-044e75296043-image.png
Please make it easier to add dictionaries yourself. All the links you shared in the main thread are broken nowadays becuase they link to https://pnlpal.dev/ instead of https://pnl.dev/.
Two days ago, I got a billing alert from Google Cloud that stopped me in my tracks.
Over 600 kr. In two days.
To put that in perspective: that's more than my entire revenue from PNL Pro for the past six months. Three days of premium TTS voices nearly wiped out the project.
Google Cloud billing showing 4140% increase
That spike? 4140% higher than last month. Not a glitch (as much as I wanted to believe), just the reality of API economics I hadn't fully understood.
How did I get here? Let me rewind a few months.
The JourneyOctober 7, 2025 — I launched PNL Pro. I was excited. The first version only had cloud syncing, and honestly, only one friend seemed to care. But it was a start.
October 25, 2025 — The first real milestone: PNL Reader 2.0. I added a read-aloud feature using Google's TTS service. Users could now listen to web novels chapter after chapter, hands-free.
November 22, 2025 — The second milestone: TTS Pro for Dictionariez. AI-powered text-to-speech for language learners to hear sentences or paragraphs.
Both products used Google's Chirp3-HD voices — premium AI voices that sound remarkably human.
The feedback was great. Users loved the quality. I was proud of what I'd shipped.
For months, everything seemed fine. The bills were manageable. Then came March 2026.
I'd made a classic startup mistake: I optimized for user experience without fully modeling the unit economics at scale.
The NumbersHere's what the billing breakdown revealed:
Cost breakdown by service
3,638,082 characters processed through Chirp3-HD in just a few days.
Google charges US$30 per million characters for Chirp3-HD. Standard voices? US$4 per million, 7.5x cheaper.
$30 per million sounds fine until you understand the use case. PNL Reader is designed for continuous reading. A user going through 50 chapters consumes ~250,000 characters per session.
The math:
250,000 characters × Chirp3-HD = $7.50 per session Same session with Standard voices = $1.00 Power user reading daily for a month = $225/month for ONE userOne user's weekend of reading could cost more than their annual subscription.
With multiple power users consuming millions of characters daily, the costs scaled faster than I'd anticipated.
The PivotOnce I understood the problem, I moved quickly.
Immediate action: Switched to Google Standard voices. Lower quality, but sustainable. This is temporary—at most until the end of this month while I implement better solutions.
What I'm exploring:
1. Local TTS. Browsers and operating systems have surprisingly good built-in voices now. Chrome, macOS, and Windows all support local TTS—free and faster than cloud APIs (no network latency).
2. Tiered pricing. Match the cost structure to usage:
Free tier: Local TTS (free, fast, decent quality) Standard tier: Google Standard voices (affordable, good quality) Premium tier: Chirp3-HD or Gemini TTS (expensive, best quality)Users who want premium voices pay for the premium cost. The economics align.
What do you think? Would you pay extra for premium voice quality? I'm genuinely curious. And your input will shape the product direction.
Looking for AlternativesIf you've worked with TTS services, I'd love your input.
What I know:
Google Chirp3-HD (expensive but great) Google Gemini TTS (great but slow) Google Standard (affordable but basic)What else is out there? ElevenLabs? Azure? Amazon Polly? OpenAI? Coqui? Tortoise TTS? If you've used any for long-form content, how's the quality, pricing, and latency?
Any recommendations would be valuable.
Key Takeaways Model your unit economics early. API costs can scale faster than you expect. Premium features need premium pricing. If the backend is expensive, the pricing should reflect that. Set up billing alerts. They won't prevent the problem, but they'll help you catch it before it's too late. What's NextThe TTS feature continues with Standard voices for now. I'm working on better solutions and expect improvements by next month.
For developers reading this: model your API costs before you ship. That premium API might be amazing, but make sure the economics work at scale.
For PNL Pro users: thanks for sticking with me. Better audio is coming.
Want to support the project? Upgrade to PNL Pro.
Dictionariez is an open-source browser extension for instant word lookup, sentence translation, realistic text-to-speech, and vocabulary building.
Translate texts, images, and audio to 100+ languages instantly. Use TextPixie’s AI translator for fast, accurate, and contextually appropriate translations.
