asu

So i've actually came back to trying to do this because killing the audio sources had unintended side effects. I had reported this under #205 but it looks like it...

I believe I have found a reliable fix/workaround to this issue. > I hacked around this by adding a static buffer of size `SAMPLE_GRANULARITY` in `soloud.cpp` and use this buffer...

> If anyone is still facing this problem #336 fixed the bug for me. The issue was that looping wave-sounds sometimes call `seek` which causes invalid memory writes, corrupting random...

I've looked into this again and I think there is no reason _not_ to use fp32 in the `arange`: 1. Checking out the `dtype` of the input tensor might not...

> Hi I am having the same issue, I am training a model performing multitasking ASR and emotion recognition, in first epoch it was showing this error, then somehow by...

I think this could be caused by an empty utterance in the dataset (maybe or maybe not due to the filtering you have done). Also, even if you don't encounter...

As a test, I replaced that part of `LatticeLmRescorer.__init__` with the following... ```python MatchType = type(_fst_spec.TableComposeOptions().table_match_type) self.old_lm_compose_cache = _fst_spec.LatticeTableComposeCache.from_compose_opts( _fst_spec.TableComposeOptions.from_matcher_opts( _fst_spec.TableMatcherOptions(), table_match_type=MatchType.MATCH_INPUT)) if not self.phi_label: self.new_lm_compose_cache = _fst_spec.LatticeTableComposeCache.from_compose_opts( _fst_spec.TableComposeOptions.from_matcher_opts( _fst_spec.TableMatcherOptions(),...

If it helps, disabling LTO for `pico-sdk/src/rp2_common/hardware_irq/irq.c` solves the "invalid offset, value too big" error in my case. The linker however then spits out a ton of undefined references related...

Looks like I got a bit further after all. In `platform.h`, I modified `WRAPPER_FUNC` to be defined as such: ```cpp #define WRAPPER_FUNC(x) __attribute__((used)) __wrap_ ## x ``` Now it successfully...

While everything seems to run fine with my workarounds applied and I have not encountered any breakage, I have realized something (which in retrospect is rather explicit in the reports...