Skip to main content

Sentence Stream

A small sentence splitter for text streams.

Sentences come out as soon as their boundary is certain, so a stream of text chunks — from an LLM, say — can be handed to a text-to-speech engine a sentence at a time instead of waiting for the whole response.

Install

pip install sentence-stream

Example

from sentence_stream import stream_to_sentences

text_chunks = [
    "Text chunks that a",
    "re not on",
    " word or se",
    "ntence boundarie",
    "s. But, they w",
    "ill sti",
    "ll get sp",
    "lit right",
    "!!! Goo",
    "d",
]

assert list(stream_to_sentences(text_chunks)) == [
    "Text chunks that are not on word or sentence boundaries.",
    "But, they will still get split right!!!",
    "Good",
]

Where the chunk boundaries fall never changes the result: the same text split into different chunks always produces the same sentences. A single complete string works too — stream_to_sentences("First sentence. Second sentence.") — for when there is nothing to stream.

For async streams, use async_stream_to_sentences:

from sentence_stream import async_stream_to_sentences

async def main(text_stream):
    async for sentence in async_stream_to_sentences(text_stream):
        print(sentence)

Pushing chunks yourself

SentenceBoundaryDetector is the same logic without the iterator, for when text arrives by callback rather than from something you can loop over. add_chunk returns the sentences that are now complete, and finish returns whatever text is left over — an empty string if there is none.

from sentence_stream import SentenceBoundaryDetector

detector = SentenceBoundaryDetector()

for sentence in detector.add_chunk("Hello world. Goodbye"):
    print(sentence)  # Hello world.

if final_sentence := detector.finish():
    print(final_sentence)  # Goodbye

A detector resets itself in finish, so it can be reused for the next stream.

What counts as a boundary

  • Terminators followed by whitespace and a capital letter: . ! ? …
  • CJK terminators, which need no whitespace after them: 。 ! ? 。, along with trailing closers such as 」 』 ) 》
  • Terminators for other scripts: ؟ (Arabic), ۔ (Urdu), । ॥ (Devanagari), ។ ៕ (Khmer), ။ (Myanmar), ། ༎ (Tibetan), ። (Ethiopic), ։ (Armenian), and ; in Greek
  • A blank line

A sentence may open with punctuation before its first letter — ¿ ¡ « „ and quotation marks — and may end with closing quotes or brackets.

Abbreviations are held rather than split on: initials (Jonas E. Smith), dotted acronyms (U.S.), and a list of known words (Mr., Mt., etc.).

Markdown emphasis is stripped from the output, so **Bold** is read as Bold. Text that only resembles markup is left alone, including 2*3*4.

Known limitations

This is a heuristic splitter, not a parser. It errs toward leaving a sentence whole, since an over-eager split is more noticeable when read aloud than a long sentence.

  • Thai and Lao are never split. Neither script has a sentence terminator — sentences are separated with a space — so there is nothing to key on.
  • Georgian is never split. The Mkhedruli script has no uppercase, and a boundary needs a capital letter after the terminator.
  • List markers are split off as their own sentence: 1. The first item becomes 1. and The first item, unless the item is emphasized.
  • A sentence-ending abbreviation stays joined to the sentence after it: I live in the U.S. How about you? comes out whole, because the same U.S. also appears mid-sentence in the U.S. Government.
  • Spaced ellipses (. . .) are treated as a boundary.

The tests/english_golden_rules.py suite records which of pySBD's golden rules pass and pins the output for those that do not.

Metadata

Release files for sentence-stream 1.4.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for sentence-stream 1.4.0
File Size Uploaded Provenance Actions
sentence_stream-1.4.0.tar.gz 24.6 kB With provenance Details for sentence_stream-1.4.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for sentence-stream 1.4.0
File Interpreter ABI Platform Actions
sentence_stream-1.4.0-py3-none-any.whl Python 3 none any Details for sentence_stream-1.4.0-py3-none-any.whl

Total release size: 39.0 kB

Release files / sentence_stream-1.4.0.tar.gz

Download URL sentence_stream-1.4.0.tar.gz
Size 24.6 kB
Tags Source
SHA-256 checksum
How to use checksums
b7c3b9c8d96090d84f07d719317e29298ad897fca666455a12160e0780890e54
BLAKE2b-256 checksum
How to use checksums
728be4b4cf9df7b9f3ceb1d0fe066243e32fb74990f2d71c668465b5e27e0001
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 28, 2026.

Transparency log

Release files / sentence_stream-1.4.0-py3-none-any.whl

Download URL sentence_stream-1.4.0-py3-none-any.whl
Size 14.4 kB
Tags Python 3
SHA-256 checksum
How to use checksums
489f008cbcaa12858088f7f83033c07a85d57b3c6a8ee7838bbfd17dd7058b4e
BLAKE2b-256 checksum
How to use checksums
8b1f46f243a1d9dfd3c4021a7d38de894babd8c96572dc035d1872a258427a8f
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 28, 2026.

Transparency log

Security for sentence-stream 1.4.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for sentence_stream-1.4.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 28, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page