Skip to content

tts word timestamps for both streaming and offline #127

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ydharavath wants to merge 2 commits into
base: main
Choose a base branch
from
Open

tts word timestamps for both streaming and offline #127

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
28edb46
feat(tts): add --word_time_offsets flag and print word timings
ydharavath Jun 22, 2026
2c28399
feat(tts): print word timestamps in streaming client + perf-client --…
ydharavath Jun 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions WORKSPACE
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,8 +76,10 @@ grpc_extra_deps()

git_repository(
name = "nvriva_common",
remote = "https://github.com/nvidia-riva/common.git",
commit = "71df98266725320a6b6b3a9f32a6da832dc93691"
# Fork of nvidia-riva/common@71df982 (current upstream main) plus the TTS
# word-timestamp proto fields (enable_word_time_offsets / WordTiming).
remote = "https://github.com/ydharavath/common.git",
commit = "560795e"
)

http_archive(
Expand Down
17 changes: 17 additions & 0 deletions riva/clients/tts/riva_tts_client.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,6 +55,9 @@ DEFINE_string(
"Input audio prompt file for Zero Shot Model. Audio length should be between 3-10 seconds.");
DEFINE_int32(zero_shot_quality, 20, "Required quality of output audio, ranges between 1-40.");
DEFINE_string(custom_dictionary, "", " User dictionary containing graph-to-phone custom words");
DEFINE_bool(
word_time_offsets, false,
"If true, request per-word start/end timestamps (returned in response metadata).");
DEFINE_string(zero_shot_transcript, "", "Transcript corresponding to Zero shot audio prompt.");
DEFINE_uint64(timeout_ms, 10000, "Timeout for GRPC channel creation");
DEFINE_uint64(max_grpc_message_size, MAX_GRPC_MESSAGE_SIZE, "Max GRPC message size");
Expand Down Expand Up @@ -239,6 +242,7 @@ main(int argc, char** argv)

request.set_sample_rate_hz(rate);
request.set_voice_name(FLAGS_voice_name);
request.set_enable_word_time_offsets(FLAGS_word_time_offsets);
if (not FLAGS_zero_shot_audio_prompt.empty()) {
auto zero_shot_data = request.mutable_zero_shot_data();
std::vector<std::shared_ptr<WaveData>> audio_prompt;
Expand Down Expand Up @@ -301,6 +305,12 @@ main(int argc, char** argv)

auto audio = response.audio();
LOG(INFO) << "Got " << audio.length() << " bytes back from server" << std::endl;
if (FLAGS_word_time_offsets) {
for (const auto& word : response.meta().words()) {
LOG(INFO) << "word: \"" << word.word() << "\" start: " << word.start_time()
<< " ms end: " << word.end_time() << " ms" << std::endl;
}
}
// Write to WAV file
if (FLAGS_audio_encoding.empty() || FLAGS_audio_encoding == "pcm") {
::riva::utils::wav::Write(
Expand Down Expand Up @@ -353,6 +363,13 @@ main(int argc, char** argv)
std::copy(opus_data, opus_data + len, std::back_inserter(opus_buffer));
audio_len += len;
}
// Word timestamps arrive in the final chunk's metadata.
if (FLAGS_word_time_offsets) {
for (const auto& word : chunk.meta().words()) {
LOG(INFO) << "word: \"" << word.word() << "\" start: " << word.start_time()
<< " ms end: " << word.end_time() << " ms" << std::endl;
}
}
}
grpc::Status rpc_status = reader->Finish();
auto end = std::chrono::steady_clock::now();
Expand Down
5 changes: 5 additions & 0 deletions riva/clients/tts/riva_tts_perf_client.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,9 @@ DEFINE_string(
language, "en-US",
"Language code as per [BCP-47](https://www.rfc-editor.org/rfc/bcp/bcp47.txt) language tag.");
DEFINE_string(voice_name, "", "Desired voice name");
DEFINE_bool(
word_time_offsets, false,
"If true, request per-word start/end timestamps (returned in response metadata).");
DEFINE_int32(num_iterations, 1, "Number of times to loop over audio files");
DEFINE_int32(num_parallel_requests, 1, "Number of parallel requests to keep in flight");
DEFINE_int32(num_sentences, 1, "Number of sentences to send");
Expand Down Expand Up @@ -153,6 +156,7 @@ synthesizeBatch(
request.set_language_code(language);
request.set_sample_rate_hz(rate);
request.set_voice_name(voice_name);
request.set_enable_word_time_offsets(FLAGS_word_time_offsets);
if (FLAGS_audio_encoding.empty() || FLAGS_audio_encoding == "pcm") {
request.set_encoding(nr::LINEAR_PCM);
} else if (FLAGS_audio_encoding == "opus") {
Expand Down Expand Up @@ -251,6 +255,7 @@ synthesizeOnline(
request.set_language_code(language);
request.set_sample_rate_hz(rate);
request.set_voice_name(voice_name);
request.set_enable_word_time_offsets(FLAGS_word_time_offsets);
auto ae = nr::AudioEncoding::ENCODING_UNSPECIFIED;
if (FLAGS_audio_encoding.empty() || FLAGS_audio_encoding == "pcm") {
ae = nr::LINEAR_PCM;
Expand Down