NLTK g2p Error: Missing averaged_perceptron_tagger_eng.weights.json (POS Tagger for "eng")

[Kush05Bhardwaj]

Hi team,

I'm trying to run the OpenVoice (or Melo TTS) pipeline, but it crashes with this error:

FileNotFoundError: [Errno 2] No such file or directory:
'C:\Users\...\nltk_data\taggers\averaged_perceptron_tagger_eng\averaged_perceptron_tagger_eng.weights.json'
It seems to be coming from the G2P component that relies on NLTK’s POS tagger, but NLTK only provides the en (not eng) version of averaged_perceptron_tagger.

Would appreciate any fix or guidance. Thanks a lot for this amazing project!

— Kushagra Bhardwaj

[dosubot[bot]]

Hi @Kush05Bhardwaj! I'm Dosu and I’m helping the OpenVoice team.

The error happens because the code is trying to load the NLTK POS tagger with the language code "eng", but NLTK only provides the resource as "en" (i.e., averaged_perceptron_tagger). This mismatch causes the missing file error.

You have two options:

1. Temporary workaround: Add this at the top of your script to try downloading the "eng" resource (works for some users, but it's not officially supported by NLTK):

   import nltk
   nltk.download('averaged_perceptron_tagger_eng')

   See user reports here.

2. Robust fix: Update the code in your environment (likely in openvoice/text/g2p_en.py) to reference "en" instead of "eng" when loading the POS tagger. This matches NLTK's naming and avoids the error. No official patch has been merged yet, but this is the root cause confirmed by code review.

This is a known issue with several user reports in the repo. Let me know if you need help finding the exact line to change.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Kush05Bhardwaj]

Hi @dosu,

Thanks for the detailed explanation. I’ve already tried both of the suggested solutions:

1. Downloaded averaged_perceptron_tagger_eng via:
   import nltk
   nltk.download('averaged_perceptron_tagger_eng')

But it failed with a FileNotFoundError since NLTK doesn’t actually provide a averaged_perceptron_tagger_eng.weights.json file.

2. Patched TAGGER_JSONS manually:

import nltk.tag.perceptron
nltk.tag.perceptron.TAGGER_JSONS["eng"] = nltk.tag.perceptron.TAGGER_JSONS["en"]

Unfortunately, this still triggers a NotImplementedError from nltk.pos_tag() because NLTK strictly checks for lang='eng' or lang='rus', and internally it defaults to loading the non-existent "eng" tagger resource.

I believe the root cause is in the TTS pipeline (likely g2p_en.py) where it’s calling pos_tag() with "eng" instead of "en". If you could help confirm the exact line or recommend the safest way to patch it locally, I’d be happy to try that.

Thanks again for supporting the community so actively!

— Kushagra Bhardwaj

[dosubot[bot]]

After a deep review, the official OpenVoice codebase does not use NLTK POS tagging or reference the 'eng' language code, so this error isn't present in the current repository. The issue is likely coming from a modified, legacy, or external component—possibly Melo TTS or a user-side patch—where nltk.pos_tag() is called with lang='eng' instead of 'en' (see similar user reports).

To fix this, search your local or external codebase for any calls to nltk.pos_tag() or similar functions using lang='eng', and change it to 'en'. If you find a code snippet and want help patching it, feel free to share it here.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Kush05Bhardwaj]

Thanks @dosu for the clarification! 🙏

You're right — it seems the issue is coming from a Melo TTS component, not OpenVoice itself. I actually came across a similar GitHub comment (screenshot attached below) where someone got it working by doing:

nltk.download('averaged_perceptron_tagger_eng')

However, in my case, this doesn't resolve the issue — the file averaged_perceptron_tagger_eng.weights.json simply isn't found, and NLTK throws a FileNotFoundError. I also tried mapping the language tag like this:

import nltk.tag.perceptron
nltk.tag.perceptron.TAGGER_JSONS["eng"] = nltk.tag.perceptron.TAGGER_JSONS["en"]

But that resulted in a NotImplementedError, likely because nltk.pos_tag() strictly checks for 'eng', which NLTK doesn't support officially.

I’ll now go through the MeloTTS-related code I’m using and search for any instance of pos_tag(..., lang='eng'), and change it to 'en' as you suggested. If I find the exact source, I’ll share it here so others facing this can patch it more easily.

Thanks again for the help!

— Kushagra Bhardwaj

[dosubot[bot]]

You're on the right track—this is a Melo TTS (not OpenVoice) issue, and the root cause is that NLTK only supports 'en' (not 'eng') for POS tagging. Downloading or patching 'averaged_perceptron_tagger_eng' won't work reliably because NLTK doesn't provide that resource, which is why you keep seeing the FileNotFoundError and NotImplementedError.

The only robust fix is to find any calls to nltk.pos_tag(..., lang='eng') in your Melo TTS code and change them to lang='en'. This matches NLTK's supported language codes and will resolve the missing file error. If you find the exact code snippet, feel free to share it here—I'm happy to review it to make sure the patch is safe and effective.

Let me know if switching to 'en' causes any other issues in your pipeline, and I'll help troubleshoot further. Sharing your findings will also help others running into the same problem.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}