Crash during live transcription in Speech-to-Text demo app

[rizalibnu]

Description

A native crash occurs during live audio transcription in the speech-to-text demo app. The app terminates silently when run via the Expo CLI, but debugging through Xcode catch a crash EXC_BAD_ACCESS on a thread. And is also replicated in the production version of Private Mind v1.1.3 currently available on the App Store using speech to text functionality inside a chat.

Steps to reproduce

1. Clone the fresh repository and checkout the main branch.
2. Navigate to apps/speech
3. Run the app on a real iOS device (yarn start and yarn:ios --device on different terminal).
4. Select Speech to Text from the demo list.
5. Tap Start Live Transcription.
6. Observe the app closing itself after approximately 1–2 minutes of continuous recording.
7. Alternative Debug Step: Run the speech demo app via Xcode. The debugger will catch a crash on thread RN_ET_Worker0: EXC_BAD_ACCESS (code=1, address=0x1220fa000) after approximately 1–2 minutes of continuous recording.

Snack or a link to a repository

No response

React Native Executorch version

main

React Native version

0.81.5

Platforms

iOS

JavaScript runtime

Hermes

Workflow

React Native

Architecture

Fabric (New Architecture)

Build type

Debug mode

Device

Real device

Device model

iPhone 11 iOS 18.0.1

AI model

WHISPER_TINY_EN

Performance logs

No response

Acknowledgements

Yes

[chmjkb]

Hi @rizalibnu, thank you for reporting this - just to make sure we can reproduce this, were you constantly talking throughout that time?

[rizalibnu]

Hi @rizalibnu, thank you for reporting this - just to make sure we can reproduce this, were you constantly talking throughout that time?

@chmjkb Yes. I simply played a random video meeting on YouTube and recorded it continuously on my phone.

[msluszniak]

Ok, I'm able to reproduce the error, however I needed to wait more like 10 minutes on my iPhone. I'm investigating this one.

[msluszniak]

Crash occurs in executorch::runtime::internal::copy_tensor_data, full backtrace from crash below:

* thread #24, name = 'RN_ET_Worker0', stop reason = EXC_BAD_ACCESS (code=1, address=0x110b6b000)

    frame #0: 0x000000020d74db24 libsystem_platform.dylib`_platform_memmove + 52

    frame #1: 0x000000010217e4c8 ExecutorchLib`executorch::runtime::internal::copy_tensor_data(executorch::runtime::etensor::Tensor const&, executorch::runtime::etensor::Tensor const&) + 112

    frame #2: 0x00000001021805a8 ExecutorchLib`executorch::runtime::Method::set_input(executorch::runtime::EValue const&, unsigned long) + 404

    frame #3: 0x000000010219dac8 ExecutorchLib`executorch::extension::module::Module::execute(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&, std::__1::vector<executorch::runtime::EValue, std::__1::allocator<executorch::runtime::EValue>> const&) + 112

    frame #4: 0x00000001061d1f40 speech.debug.dylib`executorch::extension::module::Module::forward(this=0x0000000120c9f300, input_values=size=1) at module.h:388:12

    frame #5: 0x00000001061d2658 speech.debug.dylib`executorch::extension::module::Module::forward(this=0x0000000120c9f300, input_value=0x00000001703dd340) at module.h:402:12

    frame #6: 0x00000001061d25c0 speech.debug.dylib`rnexecutorch::models::BaseModel::forward(this=0x0000000120d4d560, input_evalue=0x00000001703dd340) const at BaseModel.cpp:156:19

  * frame #7: 0x00000001061b87c8 speech.debug.dylib`rnexecutorch::models::speech_to_text::asr::ASR::encode(this=0x0000000126914ba0, waveform=size=464000) const at ASR.cpp:306:45

    frame #8: 0x00000001061b8324 speech.debug.dylib`rnexecutorch::models::speech_to_text::asr::ASR::generate(this=0x0000000126914ba0, waveform=size=464000, temperature=0, options=0x00000001703de758) const at ASR.cpp:42:44

    frame #9: 0x00000001061b94b4 speech.debug.dylib`rnexecutorch::models::speech_to_text::asr::ASR::generateWithFallback(this=0x0000000126914ba0, waveform=size=464000, options=0x00000001703de758) const at ASR.cpp:105:35

    frame #10: 0x00000001061bb0f8 speech.debug.dylib`rnexecutorch::models::speech_to_text::asr::ASR::transcribe(this=0x0000000126914ba0, waveform=size=6244801, options=0x00000001703de758) const at ASR.cpp:269:43

    frame #11: 0x00000001062654bc speech.debug.dylib`rnexecutorch::models::speech_to_text::stream::OnlineASRProcessor::processIter(this=0x00000001180817a0, options=0x00000001703de758) at OnlineASRProcessor.cpp:17:35

    frame #12: 0x00000001064a31d8 speech.debug.dylib`rnexecutorch::models::speech_to_text::SpeechToText::stream(this=0x0000000120d90fd8, callback=std::__1::shared_ptr<facebook::jsi::Function>::element_type @ 0x00000001173fe978 strong=3 weak=0, languageOption="", verbose=false) at SpeechToText.cpp:140:42

    frame #14: 0x000000010642d080 speech.debug.dylib`std::__1::invoke_result<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>, std::__1::shared_ptr<facebook::jsi::Function> const, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const, bool const>::type std::__1::invoke[abi:de200100]<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>, std::__1::shared_ptr<facebook::jsi::Function> const, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const, bool const>(__f=0x00000001703dec60, __args=std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>::element_type @ 0x0000000120d90fd8 strong=2 weak=0, __args=std::__1::shared_ptr<facebook::jsi::Function>::element_type @ 0x00000001173fe978 strong=3 weak=0, __args="", __args=0x000000011bc45f68) at invoke.h:29:10

    frame #15: 0x000000010642cff0 speech.debug.dylib`decltype(std::invoke(std::forward<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(fp), std::forward<std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>(fp), std::forward<std::__1::shared_ptr<facebook::jsi::Function> const>(fp), std::forward<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const>(fp), std::forward<bool const>(fp))) std::__1::__bind_front_op::operator()[abi:de200100]<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>, std::__1::shared_ptr<facebook::jsi::Function> const, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const, bool const>(this=0x00000001703de94f, __args=0x00000001703dec60, __args=std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>::element_type @ 0x0000000120d90fd8 strong=2 weak=0, __args=std::__1::shared_ptr<facebook::jsi::Function>::element_type @ 0x00000001173fe978 strong=3 weak=0, __args="", __args=0x000000011bc45f68) const at bind_front.h:34:12

    frame #16: 0x000000010642cfa8 speech.debug.dylib`decltype((std::__1::__bind_front_op)()(std::__1::tuple_element<0ul, std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>>::type&& std::__1::get[abi:de200100]<0ul, void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>(std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&&)(__remove_reference_t(std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&)&& std::__1::move[abi:de200100]<std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&>(std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&)((*this).__bound_args_)), std::__1::tuple_element<0ul, std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>>::type&& std::__1::get[abi:de200100]<1ul, void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>(std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&&)(__remove_reference_t(std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&)&& std::__1::move[abi:de200100]<std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&>(std::__1::tuple<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>&)((*this).__bound_args_)), std::forward<std::__1::shared_ptr<facebook::jsi::Function> const>(fp), std::forward<std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const>(fp), std::forward<bool const>(fp))) std::__1::__perfect_forward_impl<std::__1::__bind_front_op, std::__1::integer_sequence<unsigned long, 0ul, 1ul>, void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>::operator()[abi:de200100]<std::__1::shared_ptr<facebook::jsi::Function> const, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const, bool const, void>(this=0x00000001703dec60, __args=std::__1::shared_ptr<facebook::jsi::Function>::element_type @ 0x00000001173fe978 strong=3 weak=0, __args="", __args=0x000000011bc45f68) && at perfect_forward.h:78:12

    frame #18: 0x000000010642ce98 speech.debug.dylib`decltype(auto) std::__1::__apply_tuple_impl[abi:de200100]<std::__1::__bind_front_t<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>, std::__1::tuple<std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool> const, 0ul, 1ul, 2ul>(__f=0x00000001703dec60, __t=size=3, (null)=__tuple_indices<0UL, 1UL, 2UL> @ 0x00000001703de9df) at tuple:1375:5

    frame #19: 0x000000010642cb90 speech.debug.dylib`decltype(auto) std::__1::apply[abi:de200100]<std::__1::__bind_front_t<void (rnexecutorch::models::speech_to_text::SpeechToText::*)(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool), std::__1::shared_ptr<rnexecutorch::models::speech_to_text::SpeechToText>>, std::__1::tuple<std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool> const>(__f=0x00000001703dec60, __t=size=3) at tuple:1379:5

    frame #20: 0x000000010642c5f8 speech.debug.dylib`facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'()::operator()(this=0x000000011bc45f28) const at ModelHostObject.h:245:19

    frame #22: 0x000000010642c560 speech.debug.dylib`std::__1::__bind_return<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'(), std::__1::tuple<>, std::__1::tuple<>, __is_valid_bind_return<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'(), std::__1::tuple<>, std::__1::tuple<>>::value>::type std::__1::__apply_functor[abi:de200100]<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'(), std::__1::tuple<>, std::__1::tuple<>>(__f=0x000000011bc45f28, __bound_args=size=0, (null)=__tuple_indices<> @ 0x00000001703dedaf, __args=size=0) at bind.h:195:10

    frame #23: 0x000000010642c534 speech.debug.dylib`std::__1::__bind_return<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'(), std::__1::tuple<>, std::__1::tuple<>, __is_valid_bind_return<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'(), std::__1::tuple<>, std::__1::tuple<>>::value>::type std::__1::__bind<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'()>::operator()[abi:de200100]<>(this=0x000000011bc45f28) at bind.h:222:12

    frame #24: 0x000000010642c3e8 speech.debug.dylib`rnexecutorch::threads::HighPerformanceThreadPool::Task<std::__1::__bind<facebook::jsi::Value rnexecutorch::ModelHostObject<rnexecutorch::models::speech_to_text::SpeechToText>::promiseHostFunction<&rnexecutorch::models::speech_to_text::SpeechToText::stream(std::__1::shared_ptr<facebook::jsi::Function>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>, bool)>(facebook::jsi::Runtime&, facebook::jsi::Value const&, facebook::jsi::Value const*, unsigned long)::'lambda'(std::__1::shared_ptr<rnexecutorch::Promise>)::operator()(std::__1::shared_ptr<rnexecutorch::Promise>) const::'lambda'()>, void>::execute(this=0x000000011bc45f20) at HighPerformanceThreadPool.h:148:11

    frame #25: 0x000000010627fd54 speech.debug.dylib`rnexecutorch::threads::HighPerformanceThreadPool::processTask(this=0x000000011515ce00, item=0x00000001703deed8) at HighPerformanceThreadPool.h:326:18

    frame #26: 0x000000010627ed88 speech.debug.dylib`rnexecutorch::threads::HighPerformanceThreadPool::workerThread(this=0x000000011515ce00, workerIndex=0) at HighPerformanceThreadPool.h:359:7

    frame #28: 0x000000010627f7d4 speech.debug.dylib`void std::__1::__thread_execute[abi:de200100]<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct>>, void (rnexecutorch::threads::HighPerformanceThreadPool::*)(int), rnexecutorch::threads::HighPerformanceThreadPool*, unsigned long, 2ul, 3ul>(__t=size=4, (null)=__tuple_indices<2UL, 3UL> @ 0x00000001703def7f) at thread.h:205:3

    frame #29: 0x000000010627f530 speech.debug.dylib`void* std::__1::__thread_proxy[abi:de200100]<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct>>, void (rnexecutorch::threads::HighPerformanceThreadPool::*)(int), rnexecutorch::threads::HighPerformanceThreadPool*, unsigned long>>(__vp=0x000000011ae155c0) at thread.h:214:3

    frame #30: 0x000000020d7fb344 libsystem_pthread.dylib`_pthread_start + 136

[msluszniak]

@rizalibnu could you check if the fix from this PR: #805 resolves an issue for you? I was using continuously live streaming for about 30 minutes without a crash.